diff --git "a/Resources/TextEncoder.mlmodelc/model.mil" "b/Resources/TextEncoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/Resources/TextEncoder.mlmodelc/model.mil" @@ -0,0 +1,729 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.8.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})] +{ + func main(tensor input_ids) { + 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_1 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor("cast_56")]; + tensor inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = cast_1, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor("inputs_embeds_cast_fp16")]; + tensor position_embeddings_to_fp16 = const()[name = tensor("position_embeddings_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75890816)))]; + tensor input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16)[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(76009152)))]; + 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(76010752)))]; + 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 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76012352)))]; + 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(77192064)))]; + 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, x = hidden_states_1_cast_fp16)[name = tensor("linear_0_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77193664)))]; + 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(78373376)))]; + 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, x = hidden_states_1_cast_fp16)[name = tensor("linear_1_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78374976)))]; + 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(79554688)))]; + 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, x = hidden_states_1_cast_fp16)[name = tensor("linear_2_cast_fp16")]; + tensor var_113 = const()[name = tensor("op_113"), val = tensor([1, -1, 12, 64])]; + tensor var_114_cast_fp16 = reshape(shape = var_113, x = linear_0_cast_fp16)[name = tensor("op_114_cast_fp16")]; + tensor var_116 = const()[name = tensor("op_116"), val = tensor([1, -1, 12, 64])]; + tensor var_117_cast_fp16 = reshape(shape = var_116, x = linear_1_cast_fp16)[name = tensor("op_117_cast_fp16")]; + tensor var_119 = const()[name = tensor("op_119"), val = tensor([1, -1, 12, 64])]; + tensor var_120_cast_fp16 = reshape(shape = var_119, x = linear_2_cast_fp16)[name = tensor("op_120_cast_fp16")]; + tensor value_states_3_perm_0 = const()[name = tensor("value_states_3_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_17_to_fp16 = const()[name = tensor("op_17_to_fp16"), val = tensor(0x1p-3)]; + tensor mul_0_cast_fp16 = mul(x = var_114_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_0_cast_fp16")]; + tensor matmul_0_transpose_y_0 = const()[name = tensor("matmul_0_transpose_y_0"), val = tensor(true)]; + tensor matmul_0_transpose_x_0 = const()[name = tensor("matmul_0_transpose_x_0"), val = tensor(false)]; + tensor transpose_49_perm_0 = const()[name = tensor("transpose_49_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_50_perm_0 = const()[name = tensor("transpose_50_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_50 = transpose(perm = transpose_50_perm_0, x = var_117_cast_fp16)[name = tensor("transpose_118")]; + tensor transpose_49 = transpose(perm = transpose_49_perm_0, x = mul_0_cast_fp16)[name = tensor("transpose_119")]; + tensor matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_49, y = transpose_50)[name = tensor("matmul_0_cast_fp16")]; + tensor var_57_to_fp16 = const()[name = tensor("op_57_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79556288)))]; + tensor add_0_cast_fp16 = add(x = matmul_0_cast_fp16, y = var_57_to_fp16)[name = tensor("add_0_cast_fp16")]; + tensor softmax_0_axis_0 = const()[name = tensor("softmax_0_axis_0"), val = tensor(-1)]; + tensor softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = add_0_cast_fp16)[name = tensor("softmax_0_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 value_states_3_cast_fp16 = transpose(perm = value_states_3_perm_0, x = var_120_cast_fp16)[name = tensor("transpose_120")]; + tensor attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = softmax_0_cast_fp16, y = value_states_3_cast_fp16)[name = tensor("attn_output_1_cast_fp16")]; + tensor attn_output_3_perm_0 = const()[name = tensor("attn_output_3_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_124 = const()[name = tensor("op_124"), val = tensor([1, 77, 768])]; + tensor attn_output_3_cast_fp16 = transpose(perm = attn_output_3_perm_0, x = attn_output_1_cast_fp16)[name = tensor("transpose_117")]; + tensor input_5_cast_fp16 = reshape(shape = var_124, x = attn_output_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79568256)))]; + 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(80747968)))]; + 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, x = input_5_cast_fp16)[name = tensor("linear_3_cast_fp16")]; + tensor input_7_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = tensor("input_7_cast_fp16")]; + tensor input_9_axes_0 = const()[name = tensor("input_9_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(80749568)))]; + 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(80751168)))]; + tensor input_9_cast_fp16 = layer_norm(axes = input_9_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_7_cast_fp16)[name = tensor("input_9_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80752768)))]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85471424)))]; + tensor linear_4_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16, x = input_9_cast_fp16)[name = tensor("linear_4_cast_fp16")]; + tensor var_139_to_fp16 = const()[name = tensor("op_139_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_140_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_139_to_fp16)[name = tensor("op_140_cast_fp16")]; + tensor var_141_cast_fp16 = sigmoid(x = var_140_cast_fp16)[name = tensor("op_141_cast_fp16")]; + tensor input_13_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_141_cast_fp16)[name = tensor("input_13_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85477632)))]; + 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(90196288)))]; + 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, x = input_13_cast_fp16)[name = tensor("linear_5_cast_fp16")]; + tensor input_15_cast_fp16 = add(x = input_7_cast_fp16, y = linear_5_cast_fp16)[name = tensor("input_15_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(90197888)))]; + 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(90199488)))]; + 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_15_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90201088)))]; + 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(91380800)))]; + 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, x = hidden_states_7_cast_fp16)[name = tensor("linear_6_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91382400)))]; + 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(92562112)))]; + 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, x = hidden_states_7_cast_fp16)[name = tensor("linear_7_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92563712)))]; + 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(93743424)))]; + 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, x = hidden_states_7_cast_fp16)[name = tensor("linear_8_cast_fp16")]; + tensor var_171 = const()[name = tensor("op_171"), val = tensor([1, -1, 12, 64])]; + tensor var_172_cast_fp16 = reshape(shape = var_171, x = linear_6_cast_fp16)[name = tensor("op_172_cast_fp16")]; + tensor var_174 = const()[name = tensor("op_174"), val = tensor([1, -1, 12, 64])]; + tensor var_175_cast_fp16 = reshape(shape = var_174, x = linear_7_cast_fp16)[name = tensor("op_175_cast_fp16")]; + tensor var_177 = const()[name = tensor("op_177"), val = tensor([1, -1, 12, 64])]; + tensor var_178_cast_fp16 = reshape(shape = var_177, x = linear_8_cast_fp16)[name = tensor("op_178_cast_fp16")]; + tensor value_states_7_perm_0 = const()[name = tensor("value_states_7_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_1_cast_fp16 = mul(x = var_172_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_1_cast_fp16")]; + tensor matmul_1_transpose_y_0 = const()[name = tensor("matmul_1_transpose_y_0"), val = tensor(true)]; + tensor matmul_1_transpose_x_0 = const()[name = tensor("matmul_1_transpose_x_0"), val = tensor(false)]; + tensor transpose_51_perm_0 = const()[name = tensor("transpose_51_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_52_perm_0 = const()[name = tensor("transpose_52_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_52 = transpose(perm = transpose_52_perm_0, x = var_175_cast_fp16)[name = tensor("transpose_114")]; + tensor transpose_51 = transpose(perm = transpose_51_perm_0, x = mul_1_cast_fp16)[name = tensor("transpose_115")]; + tensor matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = transpose_51, y = transpose_52)[name = tensor("matmul_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = matmul_1_cast_fp16, y = var_57_to_fp16)[name = tensor("add_1_cast_fp16")]; + tensor softmax_1_axis_0 = const()[name = tensor("softmax_1_axis_0"), val = tensor(-1)]; + tensor softmax_1_cast_fp16 = softmax(axis = softmax_1_axis_0, x = add_1_cast_fp16)[name = tensor("softmax_1_cast_fp16")]; + tensor attn_output_5_transpose_x_0 = const()[name = tensor("attn_output_5_transpose_x_0"), val = tensor(false)]; + tensor attn_output_5_transpose_y_0 = const()[name = tensor("attn_output_5_transpose_y_0"), val = tensor(false)]; + tensor value_states_7_cast_fp16 = transpose(perm = value_states_7_perm_0, x = var_178_cast_fp16)[name = tensor("transpose_116")]; + tensor attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = softmax_1_cast_fp16, y = value_states_7_cast_fp16)[name = tensor("attn_output_5_cast_fp16")]; + tensor attn_output_7_perm_0 = const()[name = tensor("attn_output_7_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_182 = const()[name = tensor("op_182"), val = tensor([1, 77, 768])]; + tensor attn_output_7_cast_fp16 = transpose(perm = attn_output_7_perm_0, x = attn_output_5_cast_fp16)[name = tensor("transpose_113")]; + tensor input_17_cast_fp16 = reshape(shape = var_182, x = attn_output_7_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93745024)))]; + 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(94924736)))]; + 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, x = input_17_cast_fp16)[name = tensor("linear_9_cast_fp16")]; + tensor input_19_cast_fp16 = add(x = input_15_cast_fp16, y = linear_9_cast_fp16)[name = tensor("input_19_cast_fp16")]; + tensor input_21_axes_0 = const()[name = tensor("input_21_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(94926336)))]; + 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(94927936)))]; + tensor input_21_cast_fp16 = layer_norm(axes = input_21_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_19_cast_fp16)[name = tensor("input_21_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94929536)))]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99648192)))]; + tensor linear_10_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("linear_10_cast_fp16")]; + tensor var_197_to_fp16 = const()[name = tensor("op_197_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_198_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_197_to_fp16)[name = tensor("op_198_cast_fp16")]; + tensor var_199_cast_fp16 = sigmoid(x = var_198_cast_fp16)[name = tensor("op_199_cast_fp16")]; + tensor input_25_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_199_cast_fp16)[name = tensor("input_25_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99654400)))]; + 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(104373056)))]; + 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, x = input_25_cast_fp16)[name = tensor("linear_11_cast_fp16")]; + tensor input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_11_cast_fp16)[name = tensor("input_27_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(104374656)))]; + 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(104376256)))]; + 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_27_cast_fp16)[name = tensor("hidden_states_13_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104377856)))]; + 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(105557568)))]; + 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, x = hidden_states_13_cast_fp16)[name = tensor("linear_12_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105559168)))]; + 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(106738880)))]; + 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, x = hidden_states_13_cast_fp16)[name = tensor("linear_13_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106740480)))]; + 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(107920192)))]; + 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, x = hidden_states_13_cast_fp16)[name = tensor("linear_14_cast_fp16")]; + tensor var_229 = const()[name = tensor("op_229"), val = tensor([1, -1, 12, 64])]; + tensor var_230_cast_fp16 = reshape(shape = var_229, x = linear_12_cast_fp16)[name = tensor("op_230_cast_fp16")]; + tensor var_232 = const()[name = tensor("op_232"), val = tensor([1, -1, 12, 64])]; + tensor var_233_cast_fp16 = reshape(shape = var_232, x = linear_13_cast_fp16)[name = tensor("op_233_cast_fp16")]; + tensor var_235 = const()[name = tensor("op_235"), val = tensor([1, -1, 12, 64])]; + tensor var_236_cast_fp16 = reshape(shape = var_235, x = linear_14_cast_fp16)[name = tensor("op_236_cast_fp16")]; + tensor value_states_11_perm_0 = const()[name = tensor("value_states_11_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_2_cast_fp16 = mul(x = var_230_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_2_cast_fp16")]; + tensor matmul_2_transpose_y_0 = const()[name = tensor("matmul_2_transpose_y_0"), val = tensor(true)]; + tensor matmul_2_transpose_x_0 = const()[name = tensor("matmul_2_transpose_x_0"), val = tensor(false)]; + tensor transpose_53_perm_0 = const()[name = tensor("transpose_53_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_54_perm_0 = const()[name = tensor("transpose_54_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_54 = transpose(perm = transpose_54_perm_0, x = var_233_cast_fp16)[name = tensor("transpose_110")]; + tensor transpose_53 = transpose(perm = transpose_53_perm_0, x = mul_2_cast_fp16)[name = tensor("transpose_111")]; + tensor matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = transpose_53, y = transpose_54)[name = tensor("matmul_2_cast_fp16")]; + tensor add_2_cast_fp16 = add(x = matmul_2_cast_fp16, y = var_57_to_fp16)[name = tensor("add_2_cast_fp16")]; + tensor softmax_2_axis_0 = const()[name = tensor("softmax_2_axis_0"), val = tensor(-1)]; + tensor softmax_2_cast_fp16 = softmax(axis = softmax_2_axis_0, x = add_2_cast_fp16)[name = tensor("softmax_2_cast_fp16")]; + tensor attn_output_9_transpose_x_0 = const()[name = tensor("attn_output_9_transpose_x_0"), val = tensor(false)]; + tensor attn_output_9_transpose_y_0 = const()[name = tensor("attn_output_9_transpose_y_0"), val = tensor(false)]; + tensor value_states_11_cast_fp16 = transpose(perm = value_states_11_perm_0, x = var_236_cast_fp16)[name = tensor("transpose_112")]; + tensor attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = softmax_2_cast_fp16, y = value_states_11_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_240 = const()[name = tensor("op_240"), val = tensor([1, 77, 768])]; + tensor attn_output_11_cast_fp16 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = tensor("transpose_109")]; + tensor input_29_cast_fp16 = reshape(shape = var_240, x = attn_output_11_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107921792)))]; + 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(109101504)))]; + 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, x = input_29_cast_fp16)[name = tensor("linear_15_cast_fp16")]; + tensor input_31_cast_fp16 = add(x = input_27_cast_fp16, y = linear_15_cast_fp16)[name = tensor("input_31_cast_fp16")]; + tensor input_33_axes_0 = const()[name = tensor("input_33_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(109103104)))]; + 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(109104704)))]; + tensor input_33_cast_fp16 = layer_norm(axes = input_33_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_31_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109106304)))]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113824960)))]; + tensor linear_16_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16, x = input_33_cast_fp16)[name = tensor("linear_16_cast_fp16")]; + tensor var_255_to_fp16 = const()[name = tensor("op_255_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_256_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_255_to_fp16)[name = tensor("op_256_cast_fp16")]; + tensor var_257_cast_fp16 = sigmoid(x = var_256_cast_fp16)[name = tensor("op_257_cast_fp16")]; + tensor input_37_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_257_cast_fp16)[name = tensor("input_37_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113831168)))]; + 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(118549824)))]; + 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, x = input_37_cast_fp16)[name = tensor("linear_17_cast_fp16")]; + tensor input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_17_cast_fp16)[name = tensor("input_39_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(118551424)))]; + 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(118553024)))]; + 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_39_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118554624)))]; + 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(119734336)))]; + 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, x = hidden_states_19_cast_fp16)[name = tensor("linear_18_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119735936)))]; + 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(120915648)))]; + 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, x = hidden_states_19_cast_fp16)[name = tensor("linear_19_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120917248)))]; + 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(122096960)))]; + 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, x = hidden_states_19_cast_fp16)[name = tensor("linear_20_cast_fp16")]; + tensor var_287 = const()[name = tensor("op_287"), val = tensor([1, -1, 12, 64])]; + tensor var_288_cast_fp16 = reshape(shape = var_287, x = linear_18_cast_fp16)[name = tensor("op_288_cast_fp16")]; + tensor var_290 = const()[name = tensor("op_290"), val = tensor([1, -1, 12, 64])]; + tensor var_291_cast_fp16 = reshape(shape = var_290, x = linear_19_cast_fp16)[name = tensor("op_291_cast_fp16")]; + tensor var_293 = const()[name = tensor("op_293"), val = tensor([1, -1, 12, 64])]; + tensor var_294_cast_fp16 = reshape(shape = var_293, x = linear_20_cast_fp16)[name = tensor("op_294_cast_fp16")]; + tensor value_states_15_perm_0 = const()[name = tensor("value_states_15_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_3_cast_fp16 = mul(x = var_288_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_3_cast_fp16")]; + tensor matmul_3_transpose_y_0 = const()[name = tensor("matmul_3_transpose_y_0"), val = tensor(true)]; + tensor matmul_3_transpose_x_0 = const()[name = tensor("matmul_3_transpose_x_0"), val = tensor(false)]; + tensor transpose_55_perm_0 = const()[name = tensor("transpose_55_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_56_perm_0 = const()[name = tensor("transpose_56_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_56 = transpose(perm = transpose_56_perm_0, x = var_291_cast_fp16)[name = tensor("transpose_106")]; + tensor transpose_55 = transpose(perm = transpose_55_perm_0, x = mul_3_cast_fp16)[name = tensor("transpose_107")]; + tensor matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = transpose_55, y = transpose_56)[name = tensor("matmul_3_cast_fp16")]; + tensor add_3_cast_fp16 = add(x = matmul_3_cast_fp16, y = var_57_to_fp16)[name = tensor("add_3_cast_fp16")]; + tensor softmax_3_axis_0 = const()[name = tensor("softmax_3_axis_0"), val = tensor(-1)]; + tensor softmax_3_cast_fp16 = softmax(axis = softmax_3_axis_0, x = add_3_cast_fp16)[name = tensor("softmax_3_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 value_states_15_cast_fp16 = transpose(perm = value_states_15_perm_0, x = var_294_cast_fp16)[name = tensor("transpose_108")]; + tensor attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = softmax_3_cast_fp16, y = value_states_15_cast_fp16)[name = tensor("attn_output_13_cast_fp16")]; + tensor attn_output_15_perm_0 = const()[name = tensor("attn_output_15_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_298 = const()[name = tensor("op_298"), val = tensor([1, 77, 768])]; + tensor attn_output_15_cast_fp16 = transpose(perm = attn_output_15_perm_0, x = attn_output_13_cast_fp16)[name = tensor("transpose_105")]; + tensor input_41_cast_fp16 = reshape(shape = var_298, x = attn_output_15_cast_fp16)[name = tensor("input_41_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122098560)))]; + 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(123278272)))]; + 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, x = input_41_cast_fp16)[name = tensor("linear_21_cast_fp16")]; + tensor input_43_cast_fp16 = add(x = input_39_cast_fp16, y = linear_21_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_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(123279872)))]; + 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(123281472)))]; + tensor input_45_cast_fp16 = layer_norm(axes = input_45_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_43_cast_fp16)[name = tensor("input_45_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123283072)))]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128001728)))]; + tensor linear_22_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = tensor("linear_22_cast_fp16")]; + tensor var_313_to_fp16 = const()[name = tensor("op_313_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_314_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_313_to_fp16)[name = tensor("op_314_cast_fp16")]; + tensor var_315_cast_fp16 = sigmoid(x = var_314_cast_fp16)[name = tensor("op_315_cast_fp16")]; + tensor input_49_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_315_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128007936)))]; + 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(132726592)))]; + 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, x = input_49_cast_fp16)[name = tensor("linear_23_cast_fp16")]; + tensor input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_23_cast_fp16)[name = tensor("input_51_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(132728192)))]; + 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(132729792)))]; + 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_51_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132731392)))]; + 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(133911104)))]; + 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, x = hidden_states_25_cast_fp16)[name = tensor("linear_24_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133912704)))]; + 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(135092416)))]; + 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, x = hidden_states_25_cast_fp16)[name = tensor("linear_25_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135094016)))]; + 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(136273728)))]; + 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, x = hidden_states_25_cast_fp16)[name = tensor("linear_26_cast_fp16")]; + tensor var_345 = const()[name = tensor("op_345"), val = tensor([1, -1, 12, 64])]; + tensor var_346_cast_fp16 = reshape(shape = var_345, x = linear_24_cast_fp16)[name = tensor("op_346_cast_fp16")]; + tensor var_348 = const()[name = tensor("op_348"), val = tensor([1, -1, 12, 64])]; + tensor var_349_cast_fp16 = reshape(shape = var_348, x = linear_25_cast_fp16)[name = tensor("op_349_cast_fp16")]; + tensor var_351 = const()[name = tensor("op_351"), val = tensor([1, -1, 12, 64])]; + tensor var_352_cast_fp16 = reshape(shape = var_351, x = linear_26_cast_fp16)[name = tensor("op_352_cast_fp16")]; + tensor value_states_19_perm_0 = const()[name = tensor("value_states_19_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_4_cast_fp16 = mul(x = var_346_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_4_cast_fp16")]; + tensor matmul_4_transpose_y_0 = const()[name = tensor("matmul_4_transpose_y_0"), val = tensor(true)]; + tensor matmul_4_transpose_x_0 = const()[name = tensor("matmul_4_transpose_x_0"), val = tensor(false)]; + tensor transpose_57_perm_0 = const()[name = tensor("transpose_57_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_58_perm_0 = const()[name = tensor("transpose_58_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_58 = transpose(perm = transpose_58_perm_0, x = var_349_cast_fp16)[name = tensor("transpose_102")]; + tensor transpose_57 = transpose(perm = transpose_57_perm_0, x = mul_4_cast_fp16)[name = tensor("transpose_103")]; + tensor matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = transpose_57, y = transpose_58)[name = tensor("matmul_4_cast_fp16")]; + tensor add_4_cast_fp16 = add(x = matmul_4_cast_fp16, y = var_57_to_fp16)[name = tensor("add_4_cast_fp16")]; + tensor softmax_4_axis_0 = const()[name = tensor("softmax_4_axis_0"), val = tensor(-1)]; + tensor softmax_4_cast_fp16 = softmax(axis = softmax_4_axis_0, x = add_4_cast_fp16)[name = tensor("softmax_4_cast_fp16")]; + tensor attn_output_17_transpose_x_0 = const()[name = tensor("attn_output_17_transpose_x_0"), val = tensor(false)]; + tensor attn_output_17_transpose_y_0 = const()[name = tensor("attn_output_17_transpose_y_0"), val = tensor(false)]; + tensor value_states_19_cast_fp16 = transpose(perm = value_states_19_perm_0, x = var_352_cast_fp16)[name = tensor("transpose_104")]; + tensor attn_output_17_cast_fp16 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = softmax_4_cast_fp16, y = value_states_19_cast_fp16)[name = tensor("attn_output_17_cast_fp16")]; + tensor attn_output_19_perm_0 = const()[name = tensor("attn_output_19_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_356 = const()[name = tensor("op_356"), val = tensor([1, 77, 768])]; + tensor attn_output_19_cast_fp16 = transpose(perm = attn_output_19_perm_0, x = attn_output_17_cast_fp16)[name = tensor("transpose_101")]; + tensor input_53_cast_fp16 = reshape(shape = var_356, x = attn_output_19_cast_fp16)[name = tensor("input_53_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136275328)))]; + 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(137455040)))]; + 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, x = input_53_cast_fp16)[name = tensor("linear_27_cast_fp16")]; + tensor input_55_cast_fp16 = add(x = input_51_cast_fp16, y = linear_27_cast_fp16)[name = tensor("input_55_cast_fp16")]; + tensor input_57_axes_0 = const()[name = tensor("input_57_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(137456640)))]; + 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(137458240)))]; + tensor input_57_cast_fp16 = layer_norm(axes = input_57_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_55_cast_fp16)[name = tensor("input_57_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137459840)))]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142178496)))]; + tensor linear_28_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16, x = input_57_cast_fp16)[name = tensor("linear_28_cast_fp16")]; + tensor var_371_to_fp16 = const()[name = tensor("op_371_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_372_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_371_to_fp16)[name = tensor("op_372_cast_fp16")]; + tensor var_373_cast_fp16 = sigmoid(x = var_372_cast_fp16)[name = tensor("op_373_cast_fp16")]; + tensor input_61_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_373_cast_fp16)[name = tensor("input_61_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142184704)))]; + 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(146903360)))]; + 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, x = input_61_cast_fp16)[name = tensor("linear_29_cast_fp16")]; + tensor input_63_cast_fp16 = add(x = input_55_cast_fp16, y = linear_29_cast_fp16)[name = tensor("input_63_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(146904960)))]; + 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(146906560)))]; + 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_63_cast_fp16)[name = tensor("hidden_states_31_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146908160)))]; + 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(148087872)))]; + 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, x = hidden_states_31_cast_fp16)[name = tensor("linear_30_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148089472)))]; + 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(149269184)))]; + 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, x = hidden_states_31_cast_fp16)[name = tensor("linear_31_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149270784)))]; + 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(150450496)))]; + 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, x = hidden_states_31_cast_fp16)[name = tensor("linear_32_cast_fp16")]; + tensor var_403 = const()[name = tensor("op_403"), val = tensor([1, -1, 12, 64])]; + tensor var_404_cast_fp16 = reshape(shape = var_403, x = linear_30_cast_fp16)[name = tensor("op_404_cast_fp16")]; + tensor var_406 = const()[name = tensor("op_406"), val = tensor([1, -1, 12, 64])]; + tensor var_407_cast_fp16 = reshape(shape = var_406, x = linear_31_cast_fp16)[name = tensor("op_407_cast_fp16")]; + tensor var_409 = const()[name = tensor("op_409"), val = tensor([1, -1, 12, 64])]; + tensor var_410_cast_fp16 = reshape(shape = var_409, x = linear_32_cast_fp16)[name = tensor("op_410_cast_fp16")]; + tensor value_states_23_perm_0 = const()[name = tensor("value_states_23_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_5_cast_fp16 = mul(x = var_404_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_5_cast_fp16")]; + tensor matmul_5_transpose_y_0 = const()[name = tensor("matmul_5_transpose_y_0"), val = tensor(true)]; + tensor matmul_5_transpose_x_0 = const()[name = tensor("matmul_5_transpose_x_0"), val = tensor(false)]; + tensor transpose_59_perm_0 = const()[name = tensor("transpose_59_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_60_perm_0 = const()[name = tensor("transpose_60_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_60 = transpose(perm = transpose_60_perm_0, x = var_407_cast_fp16)[name = tensor("transpose_98")]; + tensor transpose_59 = transpose(perm = transpose_59_perm_0, x = mul_5_cast_fp16)[name = tensor("transpose_99")]; + tensor matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = transpose_59, y = transpose_60)[name = tensor("matmul_5_cast_fp16")]; + tensor add_5_cast_fp16 = add(x = matmul_5_cast_fp16, y = var_57_to_fp16)[name = tensor("add_5_cast_fp16")]; + tensor softmax_5_axis_0 = const()[name = tensor("softmax_5_axis_0"), val = tensor(-1)]; + tensor softmax_5_cast_fp16 = softmax(axis = softmax_5_axis_0, x = add_5_cast_fp16)[name = tensor("softmax_5_cast_fp16")]; + tensor attn_output_21_transpose_x_0 = const()[name = tensor("attn_output_21_transpose_x_0"), val = tensor(false)]; + tensor attn_output_21_transpose_y_0 = const()[name = tensor("attn_output_21_transpose_y_0"), val = tensor(false)]; + tensor value_states_23_cast_fp16 = transpose(perm = value_states_23_perm_0, x = var_410_cast_fp16)[name = tensor("transpose_100")]; + tensor attn_output_21_cast_fp16 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = softmax_5_cast_fp16, y = value_states_23_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_414 = const()[name = tensor("op_414"), val = tensor([1, 77, 768])]; + tensor attn_output_23_cast_fp16 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = tensor("transpose_97")]; + tensor input_65_cast_fp16 = reshape(shape = var_414, x = attn_output_23_cast_fp16)[name = tensor("input_65_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150452096)))]; + 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(151631808)))]; + 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, x = input_65_cast_fp16)[name = tensor("linear_33_cast_fp16")]; + tensor input_67_cast_fp16 = add(x = input_63_cast_fp16, y = linear_33_cast_fp16)[name = tensor("input_67_cast_fp16")]; + tensor input_69_axes_0 = const()[name = tensor("input_69_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(151633408)))]; + 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(151635008)))]; + tensor input_69_cast_fp16 = layer_norm(axes = input_69_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_67_cast_fp16)[name = tensor("input_69_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151636608)))]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156355264)))]; + tensor linear_34_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16, x = input_69_cast_fp16)[name = tensor("linear_34_cast_fp16")]; + tensor var_429_to_fp16 = const()[name = tensor("op_429_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_430_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_429_to_fp16)[name = tensor("op_430_cast_fp16")]; + tensor var_431_cast_fp16 = sigmoid(x = var_430_cast_fp16)[name = tensor("op_431_cast_fp16")]; + tensor input_73_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_431_cast_fp16)[name = tensor("input_73_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156361472)))]; + 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(161080128)))]; + 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, x = input_73_cast_fp16)[name = tensor("linear_35_cast_fp16")]; + tensor input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_35_cast_fp16)[name = tensor("input_75_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(161081728)))]; + 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(161083328)))]; + 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_75_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161084928)))]; + 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(162264640)))]; + 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, x = hidden_states_37_cast_fp16)[name = tensor("linear_36_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162266240)))]; + 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(163445952)))]; + 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, x = hidden_states_37_cast_fp16)[name = tensor("linear_37_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163447552)))]; + 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(164627264)))]; + 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, x = hidden_states_37_cast_fp16)[name = tensor("linear_38_cast_fp16")]; + tensor var_461 = const()[name = tensor("op_461"), val = tensor([1, -1, 12, 64])]; + tensor var_462_cast_fp16 = reshape(shape = var_461, x = linear_36_cast_fp16)[name = tensor("op_462_cast_fp16")]; + tensor var_464 = const()[name = tensor("op_464"), val = tensor([1, -1, 12, 64])]; + tensor var_465_cast_fp16 = reshape(shape = var_464, x = linear_37_cast_fp16)[name = tensor("op_465_cast_fp16")]; + tensor var_467 = const()[name = tensor("op_467"), val = tensor([1, -1, 12, 64])]; + tensor var_468_cast_fp16 = reshape(shape = var_467, x = linear_38_cast_fp16)[name = tensor("op_468_cast_fp16")]; + tensor value_states_27_perm_0 = const()[name = tensor("value_states_27_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_6_cast_fp16 = mul(x = var_462_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_6_cast_fp16")]; + tensor matmul_6_transpose_y_0 = const()[name = tensor("matmul_6_transpose_y_0"), val = tensor(true)]; + tensor matmul_6_transpose_x_0 = const()[name = tensor("matmul_6_transpose_x_0"), val = tensor(false)]; + tensor transpose_61_perm_0 = const()[name = tensor("transpose_61_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_62_perm_0 = const()[name = tensor("transpose_62_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_62 = transpose(perm = transpose_62_perm_0, x = var_465_cast_fp16)[name = tensor("transpose_94")]; + tensor transpose_61 = transpose(perm = transpose_61_perm_0, x = mul_6_cast_fp16)[name = tensor("transpose_95")]; + tensor matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = transpose_61, y = transpose_62)[name = tensor("matmul_6_cast_fp16")]; + tensor add_6_cast_fp16 = add(x = matmul_6_cast_fp16, y = var_57_to_fp16)[name = tensor("add_6_cast_fp16")]; + tensor softmax_6_axis_0 = const()[name = tensor("softmax_6_axis_0"), val = tensor(-1)]; + tensor softmax_6_cast_fp16 = softmax(axis = softmax_6_axis_0, x = add_6_cast_fp16)[name = tensor("softmax_6_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 value_states_27_cast_fp16 = transpose(perm = value_states_27_perm_0, x = var_468_cast_fp16)[name = tensor("transpose_96")]; + tensor attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = softmax_6_cast_fp16, y = value_states_27_cast_fp16)[name = tensor("attn_output_25_cast_fp16")]; + tensor attn_output_27_perm_0 = const()[name = tensor("attn_output_27_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_472 = const()[name = tensor("op_472"), val = tensor([1, 77, 768])]; + tensor attn_output_27_cast_fp16 = transpose(perm = attn_output_27_perm_0, x = attn_output_25_cast_fp16)[name = tensor("transpose_93")]; + tensor input_77_cast_fp16 = reshape(shape = var_472, x = attn_output_27_cast_fp16)[name = tensor("input_77_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164628864)))]; + 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(165808576)))]; + 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, x = input_77_cast_fp16)[name = tensor("linear_39_cast_fp16")]; + tensor input_79_cast_fp16 = add(x = input_75_cast_fp16, y = linear_39_cast_fp16)[name = tensor("input_79_cast_fp16")]; + tensor input_81_axes_0 = const()[name = tensor("input_81_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(165810176)))]; + 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(165811776)))]; + tensor input_81_cast_fp16 = layer_norm(axes = input_81_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_79_cast_fp16)[name = tensor("input_81_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165813376)))]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(170532032)))]; + tensor linear_40_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16, x = input_81_cast_fp16)[name = tensor("linear_40_cast_fp16")]; + tensor var_487_to_fp16 = const()[name = tensor("op_487_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_488_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_487_to_fp16)[name = tensor("op_488_cast_fp16")]; + tensor var_489_cast_fp16 = sigmoid(x = var_488_cast_fp16)[name = tensor("op_489_cast_fp16")]; + tensor input_85_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_489_cast_fp16)[name = tensor("input_85_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(170538240)))]; + 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(175256896)))]; + 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, x = input_85_cast_fp16)[name = tensor("linear_41_cast_fp16")]; + tensor input_87_cast_fp16 = add(x = input_79_cast_fp16, y = linear_41_cast_fp16)[name = tensor("input_87_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(175258496)))]; + 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(175260096)))]; + 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_87_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175261696)))]; + 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(176441408)))]; + 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, x = hidden_states_43_cast_fp16)[name = tensor("linear_42_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176443008)))]; + 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(177622720)))]; + 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, x = hidden_states_43_cast_fp16)[name = tensor("linear_43_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177624320)))]; + 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(178804032)))]; + 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, x = hidden_states_43_cast_fp16)[name = tensor("linear_44_cast_fp16")]; + tensor var_519 = const()[name = tensor("op_519"), val = tensor([1, -1, 12, 64])]; + tensor var_520_cast_fp16 = reshape(shape = var_519, x = linear_42_cast_fp16)[name = tensor("op_520_cast_fp16")]; + tensor var_522 = const()[name = tensor("op_522"), val = tensor([1, -1, 12, 64])]; + tensor var_523_cast_fp16 = reshape(shape = var_522, x = linear_43_cast_fp16)[name = tensor("op_523_cast_fp16")]; + tensor var_525 = const()[name = tensor("op_525"), val = tensor([1, -1, 12, 64])]; + tensor var_526_cast_fp16 = reshape(shape = var_525, x = linear_44_cast_fp16)[name = tensor("op_526_cast_fp16")]; + tensor value_states_31_perm_0 = const()[name = tensor("value_states_31_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_7_cast_fp16 = mul(x = var_520_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_7_cast_fp16")]; + tensor matmul_7_transpose_y_0 = const()[name = tensor("matmul_7_transpose_y_0"), val = tensor(true)]; + tensor matmul_7_transpose_x_0 = const()[name = tensor("matmul_7_transpose_x_0"), val = tensor(false)]; + tensor transpose_63_perm_0 = const()[name = tensor("transpose_63_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_64_perm_0 = const()[name = tensor("transpose_64_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_64 = transpose(perm = transpose_64_perm_0, x = var_523_cast_fp16)[name = tensor("transpose_90")]; + tensor transpose_63 = transpose(perm = transpose_63_perm_0, x = mul_7_cast_fp16)[name = tensor("transpose_91")]; + tensor matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = transpose_63, y = transpose_64)[name = tensor("matmul_7_cast_fp16")]; + tensor add_7_cast_fp16 = add(x = matmul_7_cast_fp16, y = var_57_to_fp16)[name = tensor("add_7_cast_fp16")]; + tensor softmax_7_axis_0 = const()[name = tensor("softmax_7_axis_0"), val = tensor(-1)]; + tensor softmax_7_cast_fp16 = softmax(axis = softmax_7_axis_0, x = add_7_cast_fp16)[name = tensor("softmax_7_cast_fp16")]; + tensor attn_output_29_transpose_x_0 = const()[name = tensor("attn_output_29_transpose_x_0"), val = tensor(false)]; + tensor attn_output_29_transpose_y_0 = const()[name = tensor("attn_output_29_transpose_y_0"), val = tensor(false)]; + tensor value_states_31_cast_fp16 = transpose(perm = value_states_31_perm_0, x = var_526_cast_fp16)[name = tensor("transpose_92")]; + tensor attn_output_29_cast_fp16 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = softmax_7_cast_fp16, y = value_states_31_cast_fp16)[name = tensor("attn_output_29_cast_fp16")]; + tensor attn_output_31_perm_0 = const()[name = tensor("attn_output_31_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_530 = const()[name = tensor("op_530"), val = tensor([1, 77, 768])]; + tensor attn_output_31_cast_fp16 = transpose(perm = attn_output_31_perm_0, x = attn_output_29_cast_fp16)[name = tensor("transpose_89")]; + tensor input_89_cast_fp16 = reshape(shape = var_530, x = attn_output_31_cast_fp16)[name = tensor("input_89_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178805632)))]; + 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(179985344)))]; + 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, x = input_89_cast_fp16)[name = tensor("linear_45_cast_fp16")]; + tensor input_91_cast_fp16 = add(x = input_87_cast_fp16, y = linear_45_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_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(179986944)))]; + 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(179988544)))]; + tensor input_93_cast_fp16 = layer_norm(axes = input_93_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_91_cast_fp16)[name = tensor("input_93_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179990144)))]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184708800)))]; + tensor linear_46_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16, x = input_93_cast_fp16)[name = tensor("linear_46_cast_fp16")]; + tensor var_545_to_fp16 = const()[name = tensor("op_545_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_546_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_545_to_fp16)[name = tensor("op_546_cast_fp16")]; + tensor var_547_cast_fp16 = sigmoid(x = var_546_cast_fp16)[name = tensor("op_547_cast_fp16")]; + tensor input_97_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_547_cast_fp16)[name = tensor("input_97_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184715008)))]; + 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(189433664)))]; + 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, x = input_97_cast_fp16)[name = tensor("linear_47_cast_fp16")]; + tensor input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_47_cast_fp16)[name = tensor("input_99_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(189435264)))]; + 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(189436864)))]; + 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_99_cast_fp16)[name = tensor("hidden_states_49_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189438464)))]; + 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(190618176)))]; + 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, x = hidden_states_49_cast_fp16)[name = tensor("linear_48_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190619776)))]; + 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(191799488)))]; + 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, x = hidden_states_49_cast_fp16)[name = tensor("linear_49_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(191801088)))]; + 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(192980800)))]; + 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, x = hidden_states_49_cast_fp16)[name = tensor("linear_50_cast_fp16")]; + tensor var_577 = const()[name = tensor("op_577"), val = tensor([1, -1, 12, 64])]; + tensor var_578_cast_fp16 = reshape(shape = var_577, x = linear_48_cast_fp16)[name = tensor("op_578_cast_fp16")]; + tensor var_580 = const()[name = tensor("op_580"), val = tensor([1, -1, 12, 64])]; + tensor var_581_cast_fp16 = reshape(shape = var_580, x = linear_49_cast_fp16)[name = tensor("op_581_cast_fp16")]; + tensor var_583 = const()[name = tensor("op_583"), val = tensor([1, -1, 12, 64])]; + tensor var_584_cast_fp16 = reshape(shape = var_583, x = linear_50_cast_fp16)[name = tensor("op_584_cast_fp16")]; + tensor value_states_35_perm_0 = const()[name = tensor("value_states_35_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_8_cast_fp16 = mul(x = var_578_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_8_cast_fp16")]; + tensor matmul_8_transpose_y_0 = const()[name = tensor("matmul_8_transpose_y_0"), val = tensor(true)]; + tensor matmul_8_transpose_x_0 = const()[name = tensor("matmul_8_transpose_x_0"), val = tensor(false)]; + tensor transpose_65_perm_0 = const()[name = tensor("transpose_65_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_66_perm_0 = const()[name = tensor("transpose_66_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_66 = transpose(perm = transpose_66_perm_0, x = var_581_cast_fp16)[name = tensor("transpose_86")]; + tensor transpose_65 = transpose(perm = transpose_65_perm_0, x = mul_8_cast_fp16)[name = tensor("transpose_87")]; + tensor matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = transpose_65, y = transpose_66)[name = tensor("matmul_8_cast_fp16")]; + tensor add_8_cast_fp16 = add(x = matmul_8_cast_fp16, y = var_57_to_fp16)[name = tensor("add_8_cast_fp16")]; + tensor softmax_8_axis_0 = const()[name = tensor("softmax_8_axis_0"), val = tensor(-1)]; + tensor softmax_8_cast_fp16 = softmax(axis = softmax_8_axis_0, x = add_8_cast_fp16)[name = tensor("softmax_8_cast_fp16")]; + tensor attn_output_33_transpose_x_0 = const()[name = tensor("attn_output_33_transpose_x_0"), val = tensor(false)]; + tensor attn_output_33_transpose_y_0 = const()[name = tensor("attn_output_33_transpose_y_0"), val = tensor(false)]; + tensor value_states_35_cast_fp16 = transpose(perm = value_states_35_perm_0, x = var_584_cast_fp16)[name = tensor("transpose_88")]; + tensor attn_output_33_cast_fp16 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = softmax_8_cast_fp16, y = value_states_35_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_588 = const()[name = tensor("op_588"), val = tensor([1, 77, 768])]; + tensor attn_output_35_cast_fp16 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = tensor("transpose_85")]; + tensor input_101_cast_fp16 = reshape(shape = var_588, x = attn_output_35_cast_fp16)[name = tensor("input_101_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192982400)))]; + 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(194162112)))]; + 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, x = input_101_cast_fp16)[name = tensor("linear_51_cast_fp16")]; + tensor input_103_cast_fp16 = add(x = input_99_cast_fp16, y = linear_51_cast_fp16)[name = tensor("input_103_cast_fp16")]; + tensor input_105_axes_0 = const()[name = tensor("input_105_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(194163712)))]; + 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(194165312)))]; + tensor input_105_cast_fp16 = layer_norm(axes = input_105_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_103_cast_fp16)[name = tensor("input_105_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194166912)))]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198885568)))]; + tensor linear_52_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16, x = input_105_cast_fp16)[name = tensor("linear_52_cast_fp16")]; + tensor var_603_to_fp16 = const()[name = tensor("op_603_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_604_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_603_to_fp16)[name = tensor("op_604_cast_fp16")]; + tensor var_605_cast_fp16 = sigmoid(x = var_604_cast_fp16)[name = tensor("op_605_cast_fp16")]; + tensor input_109_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_605_cast_fp16)[name = tensor("input_109_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198891776)))]; + 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(203610432)))]; + 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, x = input_109_cast_fp16)[name = tensor("linear_53_cast_fp16")]; + tensor input_111_cast_fp16 = add(x = input_103_cast_fp16, y = linear_53_cast_fp16)[name = tensor("input_111_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(203612032)))]; + 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(203613632)))]; + 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_111_cast_fp16)[name = tensor("hidden_states_55_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203615232)))]; + 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(204794944)))]; + 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, x = hidden_states_55_cast_fp16)[name = tensor("linear_54_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204796544)))]; + 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(205976256)))]; + 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, x = hidden_states_55_cast_fp16)[name = tensor("linear_55_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205977856)))]; + 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(207157568)))]; + 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, x = hidden_states_55_cast_fp16)[name = tensor("linear_56_cast_fp16")]; + tensor var_635 = const()[name = tensor("op_635"), val = tensor([1, -1, 12, 64])]; + tensor var_636_cast_fp16 = reshape(shape = var_635, x = linear_54_cast_fp16)[name = tensor("op_636_cast_fp16")]; + tensor var_638 = const()[name = tensor("op_638"), val = tensor([1, -1, 12, 64])]; + tensor var_639_cast_fp16 = reshape(shape = var_638, x = linear_55_cast_fp16)[name = tensor("op_639_cast_fp16")]; + tensor var_641 = const()[name = tensor("op_641"), val = tensor([1, -1, 12, 64])]; + tensor var_642_cast_fp16 = reshape(shape = var_641, x = linear_56_cast_fp16)[name = tensor("op_642_cast_fp16")]; + tensor value_states_39_perm_0 = const()[name = tensor("value_states_39_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_9_cast_fp16 = mul(x = var_636_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_9_cast_fp16")]; + tensor matmul_9_transpose_y_0 = const()[name = tensor("matmul_9_transpose_y_0"), val = tensor(true)]; + tensor matmul_9_transpose_x_0 = const()[name = tensor("matmul_9_transpose_x_0"), val = tensor(false)]; + tensor transpose_67_perm_0 = const()[name = tensor("transpose_67_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_68_perm_0 = const()[name = tensor("transpose_68_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_68 = transpose(perm = transpose_68_perm_0, x = var_639_cast_fp16)[name = tensor("transpose_82")]; + tensor transpose_67 = transpose(perm = transpose_67_perm_0, x = mul_9_cast_fp16)[name = tensor("transpose_83")]; + tensor matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = transpose_67, y = transpose_68)[name = tensor("matmul_9_cast_fp16")]; + tensor add_9_cast_fp16 = add(x = matmul_9_cast_fp16, y = var_57_to_fp16)[name = tensor("add_9_cast_fp16")]; + tensor softmax_9_axis_0 = const()[name = tensor("softmax_9_axis_0"), val = tensor(-1)]; + tensor softmax_9_cast_fp16 = softmax(axis = softmax_9_axis_0, x = add_9_cast_fp16)[name = tensor("softmax_9_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 value_states_39_cast_fp16 = transpose(perm = value_states_39_perm_0, x = var_642_cast_fp16)[name = tensor("transpose_84")]; + tensor attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = softmax_9_cast_fp16, y = value_states_39_cast_fp16)[name = tensor("attn_output_37_cast_fp16")]; + tensor attn_output_39_perm_0 = const()[name = tensor("attn_output_39_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_646 = const()[name = tensor("op_646"), val = tensor([1, 77, 768])]; + tensor attn_output_39_cast_fp16 = transpose(perm = attn_output_39_perm_0, x = attn_output_37_cast_fp16)[name = tensor("transpose_81")]; + tensor input_113_cast_fp16 = reshape(shape = var_646, x = attn_output_39_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207159168)))]; + 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(208338880)))]; + 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, x = input_113_cast_fp16)[name = tensor("linear_57_cast_fp16")]; + tensor input_115_cast_fp16 = add(x = input_111_cast_fp16, y = linear_57_cast_fp16)[name = tensor("input_115_cast_fp16")]; + tensor input_117_axes_0 = const()[name = tensor("input_117_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(208340480)))]; + 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(208342080)))]; + tensor input_117_cast_fp16 = layer_norm(axes = input_117_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_115_cast_fp16)[name = tensor("input_117_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208343680)))]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213062336)))]; + tensor linear_58_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16, x = input_117_cast_fp16)[name = tensor("linear_58_cast_fp16")]; + tensor var_661_to_fp16 = const()[name = tensor("op_661_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_662_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_661_to_fp16)[name = tensor("op_662_cast_fp16")]; + tensor var_663_cast_fp16 = sigmoid(x = var_662_cast_fp16)[name = tensor("op_663_cast_fp16")]; + tensor input_121_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_663_cast_fp16)[name = tensor("input_121_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213068544)))]; + 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(217787200)))]; + 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, x = input_121_cast_fp16)[name = tensor("linear_59_cast_fp16")]; + tensor input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_59_cast_fp16)[name = tensor("input_123_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(217788800)))]; + 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(217790400)))]; + 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_123_cast_fp16)[name = tensor("hidden_states_61_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217792000)))]; + 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(218971712)))]; + 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, x = hidden_states_61_cast_fp16)[name = tensor("linear_60_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218973312)))]; + 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(220153024)))]; + 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, x = hidden_states_61_cast_fp16)[name = tensor("linear_61_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220154624)))]; + 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(221334336)))]; + 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, x = hidden_states_61_cast_fp16)[name = tensor("linear_62_cast_fp16")]; + tensor var_693 = const()[name = tensor("op_693"), val = tensor([1, -1, 12, 64])]; + tensor var_694_cast_fp16 = reshape(shape = var_693, x = linear_60_cast_fp16)[name = tensor("op_694_cast_fp16")]; + tensor var_696 = const()[name = tensor("op_696"), val = tensor([1, -1, 12, 64])]; + tensor var_697_cast_fp16 = reshape(shape = var_696, x = linear_61_cast_fp16)[name = tensor("op_697_cast_fp16")]; + tensor var_699 = const()[name = tensor("op_699"), val = tensor([1, -1, 12, 64])]; + tensor var_700_cast_fp16 = reshape(shape = var_699, x = linear_62_cast_fp16)[name = tensor("op_700_cast_fp16")]; + tensor value_states_43_perm_0 = const()[name = tensor("value_states_43_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_10_cast_fp16 = mul(x = var_694_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_10_cast_fp16")]; + tensor matmul_10_transpose_y_0 = const()[name = tensor("matmul_10_transpose_y_0"), val = tensor(true)]; + tensor matmul_10_transpose_x_0 = const()[name = tensor("matmul_10_transpose_x_0"), val = tensor(false)]; + tensor transpose_69_perm_0 = const()[name = tensor("transpose_69_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_70_perm_0 = const()[name = tensor("transpose_70_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_70 = transpose(perm = transpose_70_perm_0, x = var_697_cast_fp16)[name = tensor("transpose_78")]; + tensor transpose_69 = transpose(perm = transpose_69_perm_0, x = mul_10_cast_fp16)[name = tensor("transpose_79")]; + tensor matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = transpose_69, y = transpose_70)[name = tensor("matmul_10_cast_fp16")]; + tensor add_10_cast_fp16 = add(x = matmul_10_cast_fp16, y = var_57_to_fp16)[name = tensor("add_10_cast_fp16")]; + tensor softmax_10_axis_0 = const()[name = tensor("softmax_10_axis_0"), val = tensor(-1)]; + tensor softmax_10_cast_fp16 = softmax(axis = softmax_10_axis_0, x = add_10_cast_fp16)[name = tensor("softmax_10_cast_fp16")]; + tensor attn_output_41_transpose_x_0 = const()[name = tensor("attn_output_41_transpose_x_0"), val = tensor(false)]; + tensor attn_output_41_transpose_y_0 = const()[name = tensor("attn_output_41_transpose_y_0"), val = tensor(false)]; + tensor value_states_43_cast_fp16 = transpose(perm = value_states_43_perm_0, x = var_700_cast_fp16)[name = tensor("transpose_80")]; + tensor attn_output_41_cast_fp16 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = softmax_10_cast_fp16, y = value_states_43_cast_fp16)[name = tensor("attn_output_41_cast_fp16")]; + tensor attn_output_43_perm_0 = const()[name = tensor("attn_output_43_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_704 = const()[name = tensor("op_704"), val = tensor([1, 77, 768])]; + tensor attn_output_43_cast_fp16 = transpose(perm = attn_output_43_perm_0, x = attn_output_41_cast_fp16)[name = tensor("transpose_77")]; + tensor input_125_cast_fp16 = reshape(shape = var_704, x = attn_output_43_cast_fp16)[name = tensor("input_125_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221335936)))]; + 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(222515648)))]; + 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, x = input_125_cast_fp16)[name = tensor("linear_63_cast_fp16")]; + tensor input_127_cast_fp16 = add(x = input_123_cast_fp16, y = linear_63_cast_fp16)[name = tensor("input_127_cast_fp16")]; + tensor input_129_axes_0 = const()[name = tensor("input_129_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(222517248)))]; + 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(222518848)))]; + tensor input_129_cast_fp16 = layer_norm(axes = input_129_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_127_cast_fp16)[name = tensor("input_129_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222520448)))]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227239104)))]; + tensor linear_64_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16, x = input_129_cast_fp16)[name = tensor("linear_64_cast_fp16")]; + tensor var_719_to_fp16 = const()[name = tensor("op_719_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_720_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_719_to_fp16)[name = tensor("op_720_cast_fp16")]; + tensor var_721_cast_fp16 = sigmoid(x = var_720_cast_fp16)[name = tensor("op_721_cast_fp16")]; + tensor input_133_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_721_cast_fp16)[name = tensor("input_133_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227245312)))]; + 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(231963968)))]; + 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, x = input_133_cast_fp16)[name = tensor("linear_65_cast_fp16")]; + tensor input_135_cast_fp16 = add(x = input_127_cast_fp16, y = linear_65_cast_fp16)[name = tensor("input_135_cast_fp16")]; + 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(231965568)))]; + 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(231967168)))]; + 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_135_cast_fp16)[name = tensor("hidden_states_67_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231968768)))]; + 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(233148480)))]; + 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, x = hidden_states_67_cast_fp16)[name = tensor("linear_66_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233150080)))]; + 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(234329792)))]; + 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, x = hidden_states_67_cast_fp16)[name = tensor("linear_67_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234331392)))]; + 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(235511104)))]; + 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, x = hidden_states_67_cast_fp16)[name = tensor("linear_68_cast_fp16")]; + tensor var_751 = const()[name = tensor("op_751"), val = tensor([1, -1, 12, 64])]; + tensor var_752_cast_fp16 = reshape(shape = var_751, x = linear_66_cast_fp16)[name = tensor("op_752_cast_fp16")]; + tensor var_754 = const()[name = tensor("op_754"), val = tensor([1, -1, 12, 64])]; + tensor var_755_cast_fp16 = reshape(shape = var_754, x = linear_67_cast_fp16)[name = tensor("op_755_cast_fp16")]; + tensor var_757 = const()[name = tensor("op_757"), val = tensor([1, -1, 12, 64])]; + tensor var_758_cast_fp16 = reshape(shape = var_757, x = linear_68_cast_fp16)[name = tensor("op_758_cast_fp16")]; + tensor value_states_perm_0 = const()[name = tensor("value_states_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor mul_11_cast_fp16 = mul(x = var_752_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_11_cast_fp16")]; + tensor matmul_11_transpose_y_0 = const()[name = tensor("matmul_11_transpose_y_0"), val = tensor(true)]; + tensor matmul_11_transpose_x_0 = const()[name = tensor("matmul_11_transpose_x_0"), val = tensor(false)]; + tensor transpose_71_perm_0 = const()[name = tensor("transpose_71_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_72_perm_0 = const()[name = tensor("transpose_72_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_72 = transpose(perm = transpose_72_perm_0, x = var_755_cast_fp16)[name = tensor("transpose_74")]; + tensor transpose_71 = transpose(perm = transpose_71_perm_0, x = mul_11_cast_fp16)[name = tensor("transpose_75")]; + tensor matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = transpose_71, y = transpose_72)[name = tensor("matmul_11_cast_fp16")]; + tensor add_11_cast_fp16 = add(x = matmul_11_cast_fp16, y = var_57_to_fp16)[name = tensor("add_11_cast_fp16")]; + tensor softmax_11_axis_0 = const()[name = tensor("softmax_11_axis_0"), val = tensor(-1)]; + tensor softmax_11_cast_fp16 = softmax(axis = softmax_11_axis_0, x = add_11_cast_fp16)[name = tensor("softmax_11_cast_fp16")]; + tensor attn_output_45_transpose_x_0 = const()[name = tensor("attn_output_45_transpose_x_0"), val = tensor(false)]; + tensor attn_output_45_transpose_y_0 = const()[name = tensor("attn_output_45_transpose_y_0"), val = tensor(false)]; + tensor value_states_cast_fp16 = transpose(perm = value_states_perm_0, x = var_758_cast_fp16)[name = tensor("transpose_76")]; + tensor attn_output_45_cast_fp16 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = softmax_11_cast_fp16, y = value_states_cast_fp16)[name = tensor("attn_output_45_cast_fp16")]; + tensor attn_output_perm_0 = const()[name = tensor("attn_output_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_762 = const()[name = tensor("op_762"), val = tensor([1, 77, 768])]; + tensor attn_output_cast_fp16 = transpose(perm = attn_output_perm_0, x = attn_output_45_cast_fp16)[name = tensor("transpose_73")]; + tensor input_137_cast_fp16 = reshape(shape = var_762, x = attn_output_cast_fp16)[name = tensor("input_137_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235512704)))]; + 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(236692416)))]; + 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, x = input_137_cast_fp16)[name = tensor("linear_69_cast_fp16")]; + tensor input_139_cast_fp16 = add(x = input_135_cast_fp16, y = linear_69_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_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(236694016)))]; + 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(236695616)))]; + tensor input_141_cast_fp16 = layer_norm(axes = input_141_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_139_cast_fp16)[name = tensor("input_141_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236697216)))]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241415872)))]; + tensor linear_70_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16, x = input_141_cast_fp16)[name = tensor("linear_70_cast_fp16")]; + tensor var_777_to_fp16 = const()[name = tensor("op_777_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_778_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_777_to_fp16)[name = tensor("op_778_cast_fp16")]; + tensor var_779_cast_fp16 = sigmoid(x = var_778_cast_fp16)[name = tensor("op_779_cast_fp16")]; + tensor input_145_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_779_cast_fp16)[name = tensor("input_145_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241422080)))]; + 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(246140736)))]; + 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, x = input_145_cast_fp16)[name = tensor("linear_71_cast_fp16")]; + tensor input_cast_fp16 = add(x = input_139_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(246142336)))]; + 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(246143936)))]; + 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 last_hidden_state_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("last_hidden_state_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor var_790 = const()[name = tensor("op_790"), val = tensor([0])]; + tensor var_792_axis_0 = const()[name = tensor("op_792_axis_0"), val = tensor(-1)]; + tensor var_792_keep_dims_0 = const()[name = tensor("op_792_keep_dims_0"), val = tensor(false)]; + tensor var_792 = reduce_argmax(axis = var_792_axis_0, keep_dims = var_792_keep_dims_0, x = cast_1)[name = tensor("op_792")]; + 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_790, var_792))[name = tensor("stack_0")]; + tensor var_794_transpose_batch_dims_0 = const()[name = tensor("op_794_transpose_batch_dims_0"), val = tensor(0)]; + tensor var_794_transpose_cast_fp16 = gather_nd(batch_dims = var_794_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast_fp16)[name = tensor("op_794_transpose_cast_fp16")]; + tensor var_794_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_794_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor pooled_outputs = cast(dtype = var_794_cast_fp16_to_fp32_dtype_0, x = var_794_transpose_cast_fp16)[name = tensor("cast_54")]; + tensor last_hidden_state = cast(dtype = last_hidden_state_cast_fp16_to_fp32_dtype_0, x = last_hidden_state_cast_fp16)[name = tensor("cast_55")]; + } -> (last_hidden_state, pooled_outputs); +} \ No newline at end of file