program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.4.1"}, {"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_100")]; 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(101187712)))]; 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(101345472)))]; 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(101347584)))]; 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(101349696)))]; 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(103446912)))]; 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(103449024)))]; 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(105546240)))]; 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(105548352)))]; 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(107645568)))]; 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_135 = const()[name = tensor("op_135"), val = tensor([1, -1, 16, 64])]; tensor var_136_cast_fp16 = reshape(shape = var_135, x = linear_0_cast_fp16)[name = tensor("op_136_cast_fp16")]; tensor var_138 = const()[name = tensor("op_138"), val = tensor([1, -1, 16, 64])]; tensor var_139_cast_fp16 = reshape(shape = var_138, x = linear_1_cast_fp16)[name = tensor("op_139_cast_fp16")]; tensor var_141 = const()[name = tensor("op_141"), val = tensor([1, -1, 16, 64])]; tensor var_142_cast_fp16 = reshape(shape = var_141, x = linear_2_cast_fp16)[name = tensor("op_142_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_136_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_93_perm_0 = const()[name = tensor("transpose_93_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_94_perm_0 = const()[name = tensor("transpose_94_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_94 = transpose(perm = transpose_94_perm_0, x = var_139_cast_fp16)[name = tensor("transpose_228")]; tensor transpose_93 = transpose(perm = transpose_93_perm_0, x = mul_0_cast_fp16)[name = tensor("transpose_229")]; tensor matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_93, y = transpose_94)[name = tensor("matmul_0_cast_fp16")]; tensor var_58_to_fp16 = const()[name = tensor("op_58_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107647680)))]; tensor add_0_cast_fp16 = add(x = matmul_0_cast_fp16, y = var_58_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_142_cast_fp16)[name = tensor("transpose_230")]; 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_146 = const()[name = tensor("op_146"), val = tensor([1, 77, 1024])]; tensor attn_output_3_cast_fp16 = transpose(perm = attn_output_3_perm_0, x = attn_output_1_cast_fp16)[name = tensor("transpose_227")]; tensor input_5_cast_fp16 = reshape(shape = var_146, 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(107659648)))]; 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(109756864)))]; 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(109758976)))]; 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(109761088)))]; 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(109763200)))]; 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(118151872)))]; 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 input_13_mode_0 = const()[name = tensor("input_13_mode_0"), val = tensor("EXACT")]; tensor input_13_cast_fp16 = gelu(mode = input_13_mode_0, x = linear_4_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(118160128)))]; 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(126548800)))]; 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(126550912)))]; 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(126553024)))]; 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(126555136)))]; 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(128652352)))]; 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(128654464)))]; 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(130751680)))]; 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(130753792)))]; 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(132851008)))]; 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_190 = const()[name = tensor("op_190"), val = tensor([1, -1, 16, 64])]; tensor var_191_cast_fp16 = reshape(shape = var_190, x = linear_6_cast_fp16)[name = tensor("op_191_cast_fp16")]; tensor var_193 = const()[name = tensor("op_193"), val = tensor([1, -1, 16, 64])]; tensor var_194_cast_fp16 = reshape(shape = var_193, x = linear_7_cast_fp16)[name = tensor("op_194_cast_fp16")]; tensor var_196 = const()[name = tensor("op_196"), val = tensor([1, -1, 16, 64])]; tensor var_197_cast_fp16 = reshape(shape = var_196, x = linear_8_cast_fp16)[name = tensor("op_197_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_191_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_95_perm_0 = const()[name = tensor("transpose_95_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_96_perm_0 = const()[name = tensor("transpose_96_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_96 = transpose(perm = transpose_96_perm_0, x = var_194_cast_fp16)[name = tensor("transpose_224")]; tensor transpose_95 = transpose(perm = transpose_95_perm_0, x = mul_1_cast_fp16)[name = tensor("transpose_225")]; tensor matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = transpose_95, y = transpose_96)[name = tensor("matmul_1_cast_fp16")]; tensor add_1_cast_fp16 = add(x = matmul_1_cast_fp16, y = var_58_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_197_cast_fp16)[name = tensor("transpose_226")]; 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_201 = const()[name = tensor("op_201"), val = tensor([1, 77, 1024])]; tensor attn_output_7_cast_fp16 = transpose(perm = attn_output_7_perm_0, x = attn_output_5_cast_fp16)[name = tensor("transpose_223")]; tensor input_17_cast_fp16 = reshape(shape = var_201, 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(132853120)))]; 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(134950336)))]; 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(134952448)))]; 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(134954560)))]; 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(134956672)))]; 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(143345344)))]; 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 input_25_mode_0 = const()[name = tensor("input_25_mode_0"), val = tensor("EXACT")]; tensor input_25_cast_fp16 = gelu(mode = input_25_mode_0, x = linear_10_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(143353600)))]; 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(151742272)))]; 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(151744384)))]; 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(151746496)))]; 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(151748608)))]; 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(153845824)))]; 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(153847936)))]; 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(155945152)))]; 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(155947264)))]; 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(158044480)))]; 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_245 = const()[name = tensor("op_245"), val = tensor([1, -1, 16, 64])]; tensor var_246_cast_fp16 = reshape(shape = var_245, x = linear_12_cast_fp16)[name = tensor("op_246_cast_fp16")]; tensor var_248 = const()[name = tensor("op_248"), val = tensor([1, -1, 16, 64])]; tensor var_249_cast_fp16 = reshape(shape = var_248, x = linear_13_cast_fp16)[name = tensor("op_249_cast_fp16")]; tensor var_251 = const()[name = tensor("op_251"), val = tensor([1, -1, 16, 64])]; tensor var_252_cast_fp16 = reshape(shape = var_251, x = linear_14_cast_fp16)[name = tensor("op_252_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_246_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_97_perm_0 = const()[name = tensor("transpose_97_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_98_perm_0 = const()[name = tensor("transpose_98_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_98 = transpose(perm = transpose_98_perm_0, x = var_249_cast_fp16)[name = tensor("transpose_220")]; tensor transpose_97 = transpose(perm = transpose_97_perm_0, x = mul_2_cast_fp16)[name = tensor("transpose_221")]; tensor matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = transpose_97, y = transpose_98)[name = tensor("matmul_2_cast_fp16")]; tensor add_2_cast_fp16 = add(x = matmul_2_cast_fp16, y = var_58_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_252_cast_fp16)[name = tensor("transpose_222")]; 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_256 = const()[name = tensor("op_256"), val = tensor([1, 77, 1024])]; tensor attn_output_11_cast_fp16 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = tensor("transpose_219")]; tensor input_29_cast_fp16 = reshape(shape = var_256, 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(158046592)))]; 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(160143808)))]; 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(160145920)))]; 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(160148032)))]; 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(160150144)))]; 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(168538816)))]; 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 input_37_mode_0 = const()[name = tensor("input_37_mode_0"), val = tensor("EXACT")]; tensor input_37_cast_fp16 = gelu(mode = input_37_mode_0, x = linear_16_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(168547072)))]; 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(176935744)))]; 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(176937856)))]; 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(176939968)))]; 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(176942080)))]; 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(179039296)))]; 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(179041408)))]; 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(181138624)))]; 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(181140736)))]; 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(183237952)))]; 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_300 = const()[name = tensor("op_300"), val = tensor([1, -1, 16, 64])]; tensor var_301_cast_fp16 = reshape(shape = var_300, x = linear_18_cast_fp16)[name = tensor("op_301_cast_fp16")]; tensor var_303 = const()[name = tensor("op_303"), val = tensor([1, -1, 16, 64])]; tensor var_304_cast_fp16 = reshape(shape = var_303, x = linear_19_cast_fp16)[name = tensor("op_304_cast_fp16")]; tensor var_306 = const()[name = tensor("op_306"), val = tensor([1, -1, 16, 64])]; tensor var_307_cast_fp16 = reshape(shape = var_306, x = linear_20_cast_fp16)[name = tensor("op_307_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_301_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_99_perm_0 = const()[name = tensor("transpose_99_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_100_perm_0 = const()[name = tensor("transpose_100_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_100 = transpose(perm = transpose_100_perm_0, x = var_304_cast_fp16)[name = tensor("transpose_216")]; tensor transpose_99 = transpose(perm = transpose_99_perm_0, x = mul_3_cast_fp16)[name = tensor("transpose_217")]; tensor matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = transpose_99, y = transpose_100)[name = tensor("matmul_3_cast_fp16")]; tensor add_3_cast_fp16 = add(x = matmul_3_cast_fp16, y = var_58_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_307_cast_fp16)[name = tensor("transpose_218")]; 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_311 = const()[name = tensor("op_311"), val = tensor([1, 77, 1024])]; tensor attn_output_15_cast_fp16 = transpose(perm = attn_output_15_perm_0, x = attn_output_13_cast_fp16)[name = tensor("transpose_215")]; tensor input_41_cast_fp16 = reshape(shape = var_311, 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(183240064)))]; 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(185337280)))]; 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(185339392)))]; 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(185341504)))]; 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(185343616)))]; 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(193732288)))]; 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 input_49_mode_0 = const()[name = tensor("input_49_mode_0"), val = tensor("EXACT")]; tensor input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = linear_22_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(193740544)))]; 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(202129216)))]; 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(202131328)))]; 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(202133440)))]; 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(202135552)))]; 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(204232768)))]; 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(204234880)))]; 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(206332096)))]; 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(206334208)))]; 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(208431424)))]; 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_355 = const()[name = tensor("op_355"), val = tensor([1, -1, 16, 64])]; tensor var_356_cast_fp16 = reshape(shape = var_355, x = linear_24_cast_fp16)[name = tensor("op_356_cast_fp16")]; tensor var_358 = const()[name = tensor("op_358"), val = tensor([1, -1, 16, 64])]; tensor var_359_cast_fp16 = reshape(shape = var_358, x = linear_25_cast_fp16)[name = tensor("op_359_cast_fp16")]; tensor var_361 = const()[name = tensor("op_361"), val = tensor([1, -1, 16, 64])]; tensor var_362_cast_fp16 = reshape(shape = var_361, x = linear_26_cast_fp16)[name = tensor("op_362_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_356_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_101_perm_0 = const()[name = tensor("transpose_101_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_102_perm_0 = const()[name = tensor("transpose_102_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_102 = transpose(perm = transpose_102_perm_0, x = var_359_cast_fp16)[name = tensor("transpose_212")]; tensor transpose_101 = transpose(perm = transpose_101_perm_0, x = mul_4_cast_fp16)[name = tensor("transpose_213")]; tensor matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = transpose_101, y = transpose_102)[name = tensor("matmul_4_cast_fp16")]; tensor add_4_cast_fp16 = add(x = matmul_4_cast_fp16, y = var_58_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_362_cast_fp16)[name = tensor("transpose_214")]; 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_366 = const()[name = tensor("op_366"), val = tensor([1, 77, 1024])]; tensor attn_output_19_cast_fp16 = transpose(perm = attn_output_19_perm_0, x = attn_output_17_cast_fp16)[name = tensor("transpose_211")]; tensor input_53_cast_fp16 = reshape(shape = var_366, 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(208433536)))]; 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(210530752)))]; 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(210532864)))]; 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(210534976)))]; 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(210537088)))]; 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(218925760)))]; 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 input_61_mode_0 = const()[name = tensor("input_61_mode_0"), val = tensor("EXACT")]; tensor input_61_cast_fp16 = gelu(mode = input_61_mode_0, x = linear_28_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(218934016)))]; 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(227322688)))]; 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(227324800)))]; 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(227326912)))]; 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(227329024)))]; 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(229426240)))]; 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(229428352)))]; 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(231525568)))]; 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(231527680)))]; 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(233624896)))]; 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_410 = const()[name = tensor("op_410"), val = tensor([1, -1, 16, 64])]; tensor var_411_cast_fp16 = reshape(shape = var_410, x = linear_30_cast_fp16)[name = tensor("op_411_cast_fp16")]; tensor var_413 = const()[name = tensor("op_413"), val = tensor([1, -1, 16, 64])]; tensor var_414_cast_fp16 = reshape(shape = var_413, x = linear_31_cast_fp16)[name = tensor("op_414_cast_fp16")]; tensor var_416 = const()[name = tensor("op_416"), val = tensor([1, -1, 16, 64])]; tensor var_417_cast_fp16 = reshape(shape = var_416, x = linear_32_cast_fp16)[name = tensor("op_417_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_411_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_103_perm_0 = const()[name = tensor("transpose_103_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_104_perm_0 = const()[name = tensor("transpose_104_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_104 = transpose(perm = transpose_104_perm_0, x = var_414_cast_fp16)[name = tensor("transpose_208")]; tensor transpose_103 = transpose(perm = transpose_103_perm_0, x = mul_5_cast_fp16)[name = tensor("transpose_209")]; tensor matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = transpose_103, y = transpose_104)[name = tensor("matmul_5_cast_fp16")]; tensor add_5_cast_fp16 = add(x = matmul_5_cast_fp16, y = var_58_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_417_cast_fp16)[name = tensor("transpose_210")]; 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_421 = const()[name = tensor("op_421"), val = tensor([1, 77, 1024])]; tensor attn_output_23_cast_fp16 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = tensor("transpose_207")]; tensor input_65_cast_fp16 = reshape(shape = var_421, 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(233627008)))]; 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(235724224)))]; 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(235726336)))]; 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(235728448)))]; 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(235730560)))]; 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(244119232)))]; 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 input_73_mode_0 = const()[name = tensor("input_73_mode_0"), val = tensor("EXACT")]; tensor input_73_cast_fp16 = gelu(mode = input_73_mode_0, x = linear_34_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(244127488)))]; 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(252516160)))]; 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(252518272)))]; 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(252520384)))]; 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(252522496)))]; 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(254619712)))]; 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(254621824)))]; 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(256719040)))]; 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(256721152)))]; 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(258818368)))]; 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_465 = const()[name = tensor("op_465"), val = tensor([1, -1, 16, 64])]; tensor var_466_cast_fp16 = reshape(shape = var_465, x = linear_36_cast_fp16)[name = tensor("op_466_cast_fp16")]; tensor var_468 = const()[name = tensor("op_468"), val = tensor([1, -1, 16, 64])]; tensor var_469_cast_fp16 = reshape(shape = var_468, x = linear_37_cast_fp16)[name = tensor("op_469_cast_fp16")]; tensor var_471 = const()[name = tensor("op_471"), val = tensor([1, -1, 16, 64])]; tensor var_472_cast_fp16 = reshape(shape = var_471, x = linear_38_cast_fp16)[name = tensor("op_472_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_466_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_105_perm_0 = const()[name = tensor("transpose_105_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_106_perm_0 = const()[name = tensor("transpose_106_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_106 = transpose(perm = transpose_106_perm_0, x = var_469_cast_fp16)[name = tensor("transpose_204")]; tensor transpose_105 = transpose(perm = transpose_105_perm_0, x = mul_6_cast_fp16)[name = tensor("transpose_205")]; tensor matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = transpose_105, y = transpose_106)[name = tensor("matmul_6_cast_fp16")]; tensor add_6_cast_fp16 = add(x = matmul_6_cast_fp16, y = var_58_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_472_cast_fp16)[name = tensor("transpose_206")]; 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_476 = const()[name = tensor("op_476"), val = tensor([1, 77, 1024])]; tensor attn_output_27_cast_fp16 = transpose(perm = attn_output_27_perm_0, x = attn_output_25_cast_fp16)[name = tensor("transpose_203")]; tensor input_77_cast_fp16 = reshape(shape = var_476, 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(258820480)))]; 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(260917696)))]; 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(260919808)))]; 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(260921920)))]; 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(260924032)))]; 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(269312704)))]; 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 input_85_mode_0 = const()[name = tensor("input_85_mode_0"), val = tensor("EXACT")]; tensor input_85_cast_fp16 = gelu(mode = input_85_mode_0, x = linear_40_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(269320960)))]; 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(277709632)))]; 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(277711744)))]; 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(277713856)))]; 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(277715968)))]; 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(279813184)))]; 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(279815296)))]; 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(281912512)))]; 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(281914624)))]; 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(284011840)))]; 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_520 = const()[name = tensor("op_520"), val = tensor([1, -1, 16, 64])]; tensor var_521_cast_fp16 = reshape(shape = var_520, x = linear_42_cast_fp16)[name = tensor("op_521_cast_fp16")]; tensor var_523 = const()[name = tensor("op_523"), val = tensor([1, -1, 16, 64])]; tensor var_524_cast_fp16 = reshape(shape = var_523, x = linear_43_cast_fp16)[name = tensor("op_524_cast_fp16")]; tensor var_526 = const()[name = tensor("op_526"), val = tensor([1, -1, 16, 64])]; tensor var_527_cast_fp16 = reshape(shape = var_526, x = linear_44_cast_fp16)[name = tensor("op_527_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_521_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_107_perm_0 = const()[name = tensor("transpose_107_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_108_perm_0 = const()[name = tensor("transpose_108_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_108 = transpose(perm = transpose_108_perm_0, x = var_524_cast_fp16)[name = tensor("transpose_200")]; tensor transpose_107 = transpose(perm = transpose_107_perm_0, x = mul_7_cast_fp16)[name = tensor("transpose_201")]; tensor matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = transpose_107, y = transpose_108)[name = tensor("matmul_7_cast_fp16")]; tensor add_7_cast_fp16 = add(x = matmul_7_cast_fp16, y = var_58_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_527_cast_fp16)[name = tensor("transpose_202")]; 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_531 = const()[name = tensor("op_531"), val = tensor([1, 77, 1024])]; tensor attn_output_31_cast_fp16 = transpose(perm = attn_output_31_perm_0, x = attn_output_29_cast_fp16)[name = tensor("transpose_199")]; tensor input_89_cast_fp16 = reshape(shape = var_531, 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(284013952)))]; 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(286111168)))]; 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(286113280)))]; 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(286115392)))]; 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(286117504)))]; 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(294506176)))]; 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 input_97_mode_0 = const()[name = tensor("input_97_mode_0"), val = tensor("EXACT")]; tensor input_97_cast_fp16 = gelu(mode = input_97_mode_0, x = linear_46_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(294514432)))]; 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(302903104)))]; 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(302905216)))]; 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(302907328)))]; 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(302909440)))]; 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(305006656)))]; 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(305008768)))]; 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(307105984)))]; 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(307108096)))]; 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(309205312)))]; 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_575 = const()[name = tensor("op_575"), val = tensor([1, -1, 16, 64])]; tensor var_576_cast_fp16 = reshape(shape = var_575, x = linear_48_cast_fp16)[name = tensor("op_576_cast_fp16")]; tensor var_578 = const()[name = tensor("op_578"), val = tensor([1, -1, 16, 64])]; tensor var_579_cast_fp16 = reshape(shape = var_578, x = linear_49_cast_fp16)[name = tensor("op_579_cast_fp16")]; tensor var_581 = const()[name = tensor("op_581"), val = tensor([1, -1, 16, 64])]; tensor var_582_cast_fp16 = reshape(shape = var_581, x = linear_50_cast_fp16)[name = tensor("op_582_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_576_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_109_perm_0 = const()[name = tensor("transpose_109_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_110_perm_0 = const()[name = tensor("transpose_110_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_110 = transpose(perm = transpose_110_perm_0, x = var_579_cast_fp16)[name = tensor("transpose_196")]; tensor transpose_109 = transpose(perm = transpose_109_perm_0, x = mul_8_cast_fp16)[name = tensor("transpose_197")]; tensor matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = transpose_109, y = transpose_110)[name = tensor("matmul_8_cast_fp16")]; tensor add_8_cast_fp16 = add(x = matmul_8_cast_fp16, y = var_58_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_582_cast_fp16)[name = tensor("transpose_198")]; 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_586 = const()[name = tensor("op_586"), val = tensor([1, 77, 1024])]; tensor attn_output_35_cast_fp16 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = tensor("transpose_195")]; tensor input_101_cast_fp16 = reshape(shape = var_586, 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(309207424)))]; 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(311304640)))]; 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(311306752)))]; 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(311308864)))]; 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(311310976)))]; 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(319699648)))]; 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 input_109_mode_0 = const()[name = tensor("input_109_mode_0"), val = tensor("EXACT")]; tensor input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = linear_52_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(319707904)))]; 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(328096576)))]; 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(328098688)))]; 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(328100800)))]; 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(328102912)))]; 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(330200128)))]; 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(330202240)))]; 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(332299456)))]; 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(332301568)))]; 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(334398784)))]; 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_630 = const()[name = tensor("op_630"), val = tensor([1, -1, 16, 64])]; tensor var_631_cast_fp16 = reshape(shape = var_630, x = linear_54_cast_fp16)[name = tensor("op_631_cast_fp16")]; tensor var_633 = const()[name = tensor("op_633"), val = tensor([1, -1, 16, 64])]; tensor var_634_cast_fp16 = reshape(shape = var_633, x = linear_55_cast_fp16)[name = tensor("op_634_cast_fp16")]; tensor var_636 = const()[name = tensor("op_636"), val = tensor([1, -1, 16, 64])]; tensor var_637_cast_fp16 = reshape(shape = var_636, x = linear_56_cast_fp16)[name = tensor("op_637_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_631_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_111_perm_0 = const()[name = tensor("transpose_111_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_112_perm_0 = const()[name = tensor("transpose_112_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_112 = transpose(perm = transpose_112_perm_0, x = var_634_cast_fp16)[name = tensor("transpose_192")]; tensor transpose_111 = transpose(perm = transpose_111_perm_0, x = mul_9_cast_fp16)[name = tensor("transpose_193")]; tensor matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = transpose_111, y = transpose_112)[name = tensor("matmul_9_cast_fp16")]; tensor add_9_cast_fp16 = add(x = matmul_9_cast_fp16, y = var_58_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_637_cast_fp16)[name = tensor("transpose_194")]; 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_641 = const()[name = tensor("op_641"), val = tensor([1, 77, 1024])]; tensor attn_output_39_cast_fp16 = transpose(perm = attn_output_39_perm_0, x = attn_output_37_cast_fp16)[name = tensor("transpose_191")]; tensor input_113_cast_fp16 = reshape(shape = var_641, 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(334400896)))]; 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(336498112)))]; 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(336500224)))]; 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(336502336)))]; 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(336504448)))]; 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(344893120)))]; 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 input_121_mode_0 = const()[name = tensor("input_121_mode_0"), val = tensor("EXACT")]; tensor input_121_cast_fp16 = gelu(mode = input_121_mode_0, x = linear_58_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(344901376)))]; 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(353290048)))]; 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(353292160)))]; 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(353294272)))]; 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(353296384)))]; 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(355393600)))]; 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(355395712)))]; 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(357492928)))]; 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(357495040)))]; 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(359592256)))]; 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_685 = const()[name = tensor("op_685"), val = tensor([1, -1, 16, 64])]; tensor var_686_cast_fp16 = reshape(shape = var_685, x = linear_60_cast_fp16)[name = tensor("op_686_cast_fp16")]; tensor var_688 = const()[name = tensor("op_688"), val = tensor([1, -1, 16, 64])]; tensor var_689_cast_fp16 = reshape(shape = var_688, x = linear_61_cast_fp16)[name = tensor("op_689_cast_fp16")]; tensor var_691 = const()[name = tensor("op_691"), val = tensor([1, -1, 16, 64])]; tensor var_692_cast_fp16 = reshape(shape = var_691, x = linear_62_cast_fp16)[name = tensor("op_692_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_686_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_113_perm_0 = const()[name = tensor("transpose_113_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_114_perm_0 = const()[name = tensor("transpose_114_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_114 = transpose(perm = transpose_114_perm_0, x = var_689_cast_fp16)[name = tensor("transpose_188")]; tensor transpose_113 = transpose(perm = transpose_113_perm_0, x = mul_10_cast_fp16)[name = tensor("transpose_189")]; tensor matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = transpose_113, y = transpose_114)[name = tensor("matmul_10_cast_fp16")]; tensor add_10_cast_fp16 = add(x = matmul_10_cast_fp16, y = var_58_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_692_cast_fp16)[name = tensor("transpose_190")]; 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_696 = const()[name = tensor("op_696"), val = tensor([1, 77, 1024])]; tensor attn_output_43_cast_fp16 = transpose(perm = attn_output_43_perm_0, x = attn_output_41_cast_fp16)[name = tensor("transpose_187")]; tensor input_125_cast_fp16 = reshape(shape = var_696, 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(359594368)))]; 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(361691584)))]; 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(361693696)))]; 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(361695808)))]; 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(361697920)))]; 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(370086592)))]; 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 input_133_mode_0 = const()[name = tensor("input_133_mode_0"), val = tensor("EXACT")]; tensor input_133_cast_fp16 = gelu(mode = input_133_mode_0, x = linear_64_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(370094848)))]; 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(378483520)))]; 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(378485632)))]; 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(378487744)))]; 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(378489856)))]; 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(380587072)))]; 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(380589184)))]; 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(382686400)))]; 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(382688512)))]; 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(384785728)))]; 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_740 = const()[name = tensor("op_740"), val = tensor([1, -1, 16, 64])]; tensor var_741_cast_fp16 = reshape(shape = var_740, x = linear_66_cast_fp16)[name = tensor("op_741_cast_fp16")]; tensor var_743 = const()[name = tensor("op_743"), val = tensor([1, -1, 16, 64])]; tensor var_744_cast_fp16 = reshape(shape = var_743, x = linear_67_cast_fp16)[name = tensor("op_744_cast_fp16")]; tensor var_746 = const()[name = tensor("op_746"), val = tensor([1, -1, 16, 64])]; tensor var_747_cast_fp16 = reshape(shape = var_746, x = linear_68_cast_fp16)[name = tensor("op_747_cast_fp16")]; tensor value_states_47_perm_0 = const()[name = tensor("value_states_47_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_11_cast_fp16 = mul(x = var_741_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_115_perm_0 = const()[name = tensor("transpose_115_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_116_perm_0 = const()[name = tensor("transpose_116_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_116 = transpose(perm = transpose_116_perm_0, x = var_744_cast_fp16)[name = tensor("transpose_184")]; tensor transpose_115 = transpose(perm = transpose_115_perm_0, x = mul_11_cast_fp16)[name = tensor("transpose_185")]; tensor matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = transpose_115, y = transpose_116)[name = tensor("matmul_11_cast_fp16")]; tensor add_11_cast_fp16 = add(x = matmul_11_cast_fp16, y = var_58_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_47_cast_fp16 = transpose(perm = value_states_47_perm_0, x = var_747_cast_fp16)[name = tensor("transpose_186")]; 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_47_cast_fp16)[name = tensor("attn_output_45_cast_fp16")]; tensor attn_output_47_perm_0 = const()[name = tensor("attn_output_47_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_751 = const()[name = tensor("op_751"), val = tensor([1, 77, 1024])]; tensor attn_output_47_cast_fp16 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast_fp16)[name = tensor("transpose_183")]; tensor input_137_cast_fp16 = reshape(shape = var_751, x = attn_output_47_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(384787840)))]; 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(386885056)))]; 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(386887168)))]; 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(386889280)))]; 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(386891392)))]; 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(395280064)))]; 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 input_145_mode_0 = const()[name = tensor("input_145_mode_0"), val = tensor("EXACT")]; tensor input_145_cast_fp16 = gelu(mode = input_145_mode_0, x = linear_70_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(395288320)))]; 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(403676992)))]; 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_147_cast_fp16 = add(x = input_139_cast_fp16, y = linear_71_cast_fp16)[name = tensor("input_147_cast_fp16")]; tensor hidden_states_73_axes_0 = const()[name = tensor("hidden_states_73_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403679104)))]; tensor text_encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403681216)))]; tensor hidden_states_73_cast_fp16 = layer_norm(axes = hidden_states_73_axes_0, beta = text_encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor("hidden_states_73_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403683328)))]; tensor text_encoder_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405780544)))]; tensor linear_72_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor("linear_72_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405782656)))]; tensor text_encoder_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407879872)))]; tensor linear_73_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor("linear_73_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407881984)))]; tensor text_encoder_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409979200)))]; tensor linear_74_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor("linear_74_cast_fp16")]; tensor var_795 = const()[name = tensor("op_795"), val = tensor([1, -1, 16, 64])]; tensor var_796_cast_fp16 = reshape(shape = var_795, x = linear_72_cast_fp16)[name = tensor("op_796_cast_fp16")]; tensor var_798 = const()[name = tensor("op_798"), val = tensor([1, -1, 16, 64])]; tensor var_799_cast_fp16 = reshape(shape = var_798, x = linear_73_cast_fp16)[name = tensor("op_799_cast_fp16")]; tensor var_801 = const()[name = tensor("op_801"), val = tensor([1, -1, 16, 64])]; tensor var_802_cast_fp16 = reshape(shape = var_801, x = linear_74_cast_fp16)[name = tensor("op_802_cast_fp16")]; tensor value_states_51_perm_0 = const()[name = tensor("value_states_51_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_12_cast_fp16 = mul(x = var_796_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_12_cast_fp16")]; tensor matmul_12_transpose_y_0 = const()[name = tensor("matmul_12_transpose_y_0"), val = tensor(true)]; tensor matmul_12_transpose_x_0 = const()[name = tensor("matmul_12_transpose_x_0"), val = tensor(false)]; tensor transpose_117_perm_0 = const()[name = tensor("transpose_117_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_118_perm_0 = const()[name = tensor("transpose_118_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_118 = transpose(perm = transpose_118_perm_0, x = var_799_cast_fp16)[name = tensor("transpose_180")]; tensor transpose_117 = transpose(perm = transpose_117_perm_0, x = mul_12_cast_fp16)[name = tensor("transpose_181")]; tensor matmul_12_cast_fp16 = matmul(transpose_x = matmul_12_transpose_x_0, transpose_y = matmul_12_transpose_y_0, x = transpose_117, y = transpose_118)[name = tensor("matmul_12_cast_fp16")]; tensor add_12_cast_fp16 = add(x = matmul_12_cast_fp16, y = var_58_to_fp16)[name = tensor("add_12_cast_fp16")]; tensor softmax_12_axis_0 = const()[name = tensor("softmax_12_axis_0"), val = tensor(-1)]; tensor softmax_12_cast_fp16 = softmax(axis = softmax_12_axis_0, x = add_12_cast_fp16)[name = tensor("softmax_12_cast_fp16")]; tensor attn_output_49_transpose_x_0 = const()[name = tensor("attn_output_49_transpose_x_0"), val = tensor(false)]; tensor attn_output_49_transpose_y_0 = const()[name = tensor("attn_output_49_transpose_y_0"), val = tensor(false)]; tensor value_states_51_cast_fp16 = transpose(perm = value_states_51_perm_0, x = var_802_cast_fp16)[name = tensor("transpose_182")]; tensor attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = softmax_12_cast_fp16, y = value_states_51_cast_fp16)[name = tensor("attn_output_49_cast_fp16")]; tensor attn_output_51_perm_0 = const()[name = tensor("attn_output_51_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_806 = const()[name = tensor("op_806"), val = tensor([1, 77, 1024])]; tensor attn_output_51_cast_fp16 = transpose(perm = attn_output_51_perm_0, x = attn_output_49_cast_fp16)[name = tensor("transpose_179")]; tensor input_149_cast_fp16 = reshape(shape = var_806, x = attn_output_51_cast_fp16)[name = tensor("input_149_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409981312)))]; tensor text_encoder_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412078528)))]; tensor linear_75_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16, x = input_149_cast_fp16)[name = tensor("linear_75_cast_fp16")]; tensor input_151_cast_fp16 = add(x = input_147_cast_fp16, y = linear_75_cast_fp16)[name = tensor("input_151_cast_fp16")]; tensor input_153_axes_0 = const()[name = tensor("input_153_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412080640)))]; tensor text_encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412082752)))]; tensor input_153_cast_fp16 = layer_norm(axes = input_153_axes_0, beta = text_encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16, x = input_151_cast_fp16)[name = tensor("input_153_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412084864)))]; tensor text_encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420473536)))]; tensor linear_76_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16, x = input_153_cast_fp16)[name = tensor("linear_76_cast_fp16")]; tensor input_157_mode_0 = const()[name = tensor("input_157_mode_0"), val = tensor("EXACT")]; tensor input_157_cast_fp16 = gelu(mode = input_157_mode_0, x = linear_76_cast_fp16)[name = tensor("input_157_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420481792)))]; tensor text_encoder_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428870464)))]; tensor linear_77_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16, x = input_157_cast_fp16)[name = tensor("linear_77_cast_fp16")]; tensor input_159_cast_fp16 = add(x = input_151_cast_fp16, y = linear_77_cast_fp16)[name = tensor("input_159_cast_fp16")]; tensor hidden_states_79_axes_0 = const()[name = tensor("hidden_states_79_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428872576)))]; tensor text_encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428874688)))]; tensor hidden_states_79_cast_fp16 = layer_norm(axes = hidden_states_79_axes_0, beta = text_encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16, x = input_159_cast_fp16)[name = tensor("hidden_states_79_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428876800)))]; tensor text_encoder_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430974016)))]; tensor linear_78_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor("linear_78_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430976128)))]; tensor text_encoder_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433073344)))]; tensor linear_79_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor("linear_79_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433075456)))]; tensor text_encoder_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(435172672)))]; tensor linear_80_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor("linear_80_cast_fp16")]; tensor var_850 = const()[name = tensor("op_850"), val = tensor([1, -1, 16, 64])]; tensor var_851_cast_fp16 = reshape(shape = var_850, x = linear_78_cast_fp16)[name = tensor("op_851_cast_fp16")]; tensor var_853 = const()[name = tensor("op_853"), val = tensor([1, -1, 16, 64])]; tensor var_854_cast_fp16 = reshape(shape = var_853, x = linear_79_cast_fp16)[name = tensor("op_854_cast_fp16")]; tensor var_856 = const()[name = tensor("op_856"), val = tensor([1, -1, 16, 64])]; tensor var_857_cast_fp16 = reshape(shape = var_856, x = linear_80_cast_fp16)[name = tensor("op_857_cast_fp16")]; tensor value_states_55_perm_0 = const()[name = tensor("value_states_55_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_13_cast_fp16 = mul(x = var_851_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_13_cast_fp16")]; tensor matmul_13_transpose_y_0 = const()[name = tensor("matmul_13_transpose_y_0"), val = tensor(true)]; tensor matmul_13_transpose_x_0 = const()[name = tensor("matmul_13_transpose_x_0"), val = tensor(false)]; tensor transpose_119_perm_0 = const()[name = tensor("transpose_119_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_120_perm_0 = const()[name = tensor("transpose_120_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_120 = transpose(perm = transpose_120_perm_0, x = var_854_cast_fp16)[name = tensor("transpose_176")]; tensor transpose_119 = transpose(perm = transpose_119_perm_0, x = mul_13_cast_fp16)[name = tensor("transpose_177")]; tensor matmul_13_cast_fp16 = matmul(transpose_x = matmul_13_transpose_x_0, transpose_y = matmul_13_transpose_y_0, x = transpose_119, y = transpose_120)[name = tensor("matmul_13_cast_fp16")]; tensor add_13_cast_fp16 = add(x = matmul_13_cast_fp16, y = var_58_to_fp16)[name = tensor("add_13_cast_fp16")]; tensor softmax_13_axis_0 = const()[name = tensor("softmax_13_axis_0"), val = tensor(-1)]; tensor softmax_13_cast_fp16 = softmax(axis = softmax_13_axis_0, x = add_13_cast_fp16)[name = tensor("softmax_13_cast_fp16")]; tensor attn_output_53_transpose_x_0 = const()[name = tensor("attn_output_53_transpose_x_0"), val = tensor(false)]; tensor attn_output_53_transpose_y_0 = const()[name = tensor("attn_output_53_transpose_y_0"), val = tensor(false)]; tensor value_states_55_cast_fp16 = transpose(perm = value_states_55_perm_0, x = var_857_cast_fp16)[name = tensor("transpose_178")]; tensor attn_output_53_cast_fp16 = matmul(transpose_x = attn_output_53_transpose_x_0, transpose_y = attn_output_53_transpose_y_0, x = softmax_13_cast_fp16, y = value_states_55_cast_fp16)[name = tensor("attn_output_53_cast_fp16")]; tensor attn_output_55_perm_0 = const()[name = tensor("attn_output_55_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_861 = const()[name = tensor("op_861"), val = tensor([1, 77, 1024])]; tensor attn_output_55_cast_fp16 = transpose(perm = attn_output_55_perm_0, x = attn_output_53_cast_fp16)[name = tensor("transpose_175")]; tensor input_161_cast_fp16 = reshape(shape = var_861, x = attn_output_55_cast_fp16)[name = tensor("input_161_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(435174784)))]; tensor text_encoder_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(437272000)))]; tensor linear_81_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16, x = input_161_cast_fp16)[name = tensor("linear_81_cast_fp16")]; tensor input_163_cast_fp16 = add(x = input_159_cast_fp16, y = linear_81_cast_fp16)[name = tensor("input_163_cast_fp16")]; tensor input_165_axes_0 = const()[name = tensor("input_165_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(437274112)))]; tensor text_encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(437276224)))]; tensor input_165_cast_fp16 = layer_norm(axes = input_165_axes_0, beta = text_encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16, x = input_163_cast_fp16)[name = tensor("input_165_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(437278336)))]; tensor text_encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(445667008)))]; tensor linear_82_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16, x = input_165_cast_fp16)[name = tensor("linear_82_cast_fp16")]; tensor input_169_mode_0 = const()[name = tensor("input_169_mode_0"), val = tensor("EXACT")]; tensor input_169_cast_fp16 = gelu(mode = input_169_mode_0, x = linear_82_cast_fp16)[name = tensor("input_169_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(445675264)))]; tensor text_encoder_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454063936)))]; tensor linear_83_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16, x = input_169_cast_fp16)[name = tensor("linear_83_cast_fp16")]; tensor input_171_cast_fp16 = add(x = input_163_cast_fp16, y = linear_83_cast_fp16)[name = tensor("input_171_cast_fp16")]; tensor hidden_states_85_axes_0 = const()[name = tensor("hidden_states_85_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454066048)))]; tensor text_encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454068160)))]; tensor hidden_states_85_cast_fp16 = layer_norm(axes = hidden_states_85_axes_0, beta = text_encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16, x = input_171_cast_fp16)[name = tensor("hidden_states_85_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454070272)))]; tensor text_encoder_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456167488)))]; tensor linear_84_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor("linear_84_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456169600)))]; tensor text_encoder_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(458266816)))]; tensor linear_85_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor("linear_85_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(458268928)))]; tensor text_encoder_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460366144)))]; tensor linear_86_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor("linear_86_cast_fp16")]; tensor var_905 = const()[name = tensor("op_905"), val = tensor([1, -1, 16, 64])]; tensor var_906_cast_fp16 = reshape(shape = var_905, x = linear_84_cast_fp16)[name = tensor("op_906_cast_fp16")]; tensor var_908 = const()[name = tensor("op_908"), val = tensor([1, -1, 16, 64])]; tensor var_909_cast_fp16 = reshape(shape = var_908, x = linear_85_cast_fp16)[name = tensor("op_909_cast_fp16")]; tensor var_911 = const()[name = tensor("op_911"), val = tensor([1, -1, 16, 64])]; tensor var_912_cast_fp16 = reshape(shape = var_911, x = linear_86_cast_fp16)[name = tensor("op_912_cast_fp16")]; tensor value_states_59_perm_0 = const()[name = tensor("value_states_59_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_14_cast_fp16 = mul(x = var_906_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_14_cast_fp16")]; tensor matmul_14_transpose_y_0 = const()[name = tensor("matmul_14_transpose_y_0"), val = tensor(true)]; tensor matmul_14_transpose_x_0 = const()[name = tensor("matmul_14_transpose_x_0"), val = tensor(false)]; tensor transpose_121_perm_0 = const()[name = tensor("transpose_121_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_122_perm_0 = const()[name = tensor("transpose_122_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_122 = transpose(perm = transpose_122_perm_0, x = var_909_cast_fp16)[name = tensor("transpose_172")]; tensor transpose_121 = transpose(perm = transpose_121_perm_0, x = mul_14_cast_fp16)[name = tensor("transpose_173")]; tensor matmul_14_cast_fp16 = matmul(transpose_x = matmul_14_transpose_x_0, transpose_y = matmul_14_transpose_y_0, x = transpose_121, y = transpose_122)[name = tensor("matmul_14_cast_fp16")]; tensor add_14_cast_fp16 = add(x = matmul_14_cast_fp16, y = var_58_to_fp16)[name = tensor("add_14_cast_fp16")]; tensor softmax_14_axis_0 = const()[name = tensor("softmax_14_axis_0"), val = tensor(-1)]; tensor softmax_14_cast_fp16 = softmax(axis = softmax_14_axis_0, x = add_14_cast_fp16)[name = tensor("softmax_14_cast_fp16")]; tensor attn_output_57_transpose_x_0 = const()[name = tensor("attn_output_57_transpose_x_0"), val = tensor(false)]; tensor attn_output_57_transpose_y_0 = const()[name = tensor("attn_output_57_transpose_y_0"), val = tensor(false)]; tensor value_states_59_cast_fp16 = transpose(perm = value_states_59_perm_0, x = var_912_cast_fp16)[name = tensor("transpose_174")]; tensor attn_output_57_cast_fp16 = matmul(transpose_x = attn_output_57_transpose_x_0, transpose_y = attn_output_57_transpose_y_0, x = softmax_14_cast_fp16, y = value_states_59_cast_fp16)[name = tensor("attn_output_57_cast_fp16")]; tensor attn_output_59_perm_0 = const()[name = tensor("attn_output_59_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_916 = const()[name = tensor("op_916"), val = tensor([1, 77, 1024])]; tensor attn_output_59_cast_fp16 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast_fp16)[name = tensor("transpose_171")]; tensor input_173_cast_fp16 = reshape(shape = var_916, x = attn_output_59_cast_fp16)[name = tensor("input_173_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460368256)))]; tensor text_encoder_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(462465472)))]; tensor linear_87_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16, x = input_173_cast_fp16)[name = tensor("linear_87_cast_fp16")]; tensor input_175_cast_fp16 = add(x = input_171_cast_fp16, y = linear_87_cast_fp16)[name = tensor("input_175_cast_fp16")]; tensor input_177_axes_0 = const()[name = tensor("input_177_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(462467584)))]; tensor text_encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(462469696)))]; tensor input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = text_encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16, x = input_175_cast_fp16)[name = tensor("input_177_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(462471808)))]; tensor text_encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470860480)))]; tensor linear_88_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16, x = input_177_cast_fp16)[name = tensor("linear_88_cast_fp16")]; tensor input_181_mode_0 = const()[name = tensor("input_181_mode_0"), val = tensor("EXACT")]; tensor input_181_cast_fp16 = gelu(mode = input_181_mode_0, x = linear_88_cast_fp16)[name = tensor("input_181_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(470868736)))]; tensor text_encoder_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(479257408)))]; tensor linear_89_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16, x = input_181_cast_fp16)[name = tensor("linear_89_cast_fp16")]; tensor input_183_cast_fp16 = add(x = input_175_cast_fp16, y = linear_89_cast_fp16)[name = tensor("input_183_cast_fp16")]; tensor hidden_states_91_axes_0 = const()[name = tensor("hidden_states_91_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(479259520)))]; tensor text_encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(479261632)))]; tensor hidden_states_91_cast_fp16 = layer_norm(axes = hidden_states_91_axes_0, beta = text_encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16, x = input_183_cast_fp16)[name = tensor("hidden_states_91_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(479263744)))]; tensor text_encoder_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(481360960)))]; tensor linear_90_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor("linear_90_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(481363072)))]; tensor text_encoder_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(483460288)))]; tensor linear_91_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor("linear_91_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(483462400)))]; tensor text_encoder_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(485559616)))]; tensor linear_92_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor("linear_92_cast_fp16")]; tensor var_960 = const()[name = tensor("op_960"), val = tensor([1, -1, 16, 64])]; tensor var_961_cast_fp16 = reshape(shape = var_960, x = linear_90_cast_fp16)[name = tensor("op_961_cast_fp16")]; tensor var_963 = const()[name = tensor("op_963"), val = tensor([1, -1, 16, 64])]; tensor var_964_cast_fp16 = reshape(shape = var_963, x = linear_91_cast_fp16)[name = tensor("op_964_cast_fp16")]; tensor var_966 = const()[name = tensor("op_966"), val = tensor([1, -1, 16, 64])]; tensor var_967_cast_fp16 = reshape(shape = var_966, x = linear_92_cast_fp16)[name = tensor("op_967_cast_fp16")]; tensor value_states_63_perm_0 = const()[name = tensor("value_states_63_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_15_cast_fp16 = mul(x = var_961_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_15_cast_fp16")]; tensor matmul_15_transpose_y_0 = const()[name = tensor("matmul_15_transpose_y_0"), val = tensor(true)]; tensor matmul_15_transpose_x_0 = const()[name = tensor("matmul_15_transpose_x_0"), val = tensor(false)]; tensor transpose_123_perm_0 = const()[name = tensor("transpose_123_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_124_perm_0 = const()[name = tensor("transpose_124_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_124 = transpose(perm = transpose_124_perm_0, x = var_964_cast_fp16)[name = tensor("transpose_168")]; tensor transpose_123 = transpose(perm = transpose_123_perm_0, x = mul_15_cast_fp16)[name = tensor("transpose_169")]; tensor matmul_15_cast_fp16 = matmul(transpose_x = matmul_15_transpose_x_0, transpose_y = matmul_15_transpose_y_0, x = transpose_123, y = transpose_124)[name = tensor("matmul_15_cast_fp16")]; tensor add_15_cast_fp16 = add(x = matmul_15_cast_fp16, y = var_58_to_fp16)[name = tensor("add_15_cast_fp16")]; tensor softmax_15_axis_0 = const()[name = tensor("softmax_15_axis_0"), val = tensor(-1)]; tensor softmax_15_cast_fp16 = softmax(axis = softmax_15_axis_0, x = add_15_cast_fp16)[name = tensor("softmax_15_cast_fp16")]; tensor attn_output_61_transpose_x_0 = const()[name = tensor("attn_output_61_transpose_x_0"), val = tensor(false)]; tensor attn_output_61_transpose_y_0 = const()[name = tensor("attn_output_61_transpose_y_0"), val = tensor(false)]; tensor value_states_63_cast_fp16 = transpose(perm = value_states_63_perm_0, x = var_967_cast_fp16)[name = tensor("transpose_170")]; tensor attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = softmax_15_cast_fp16, y = value_states_63_cast_fp16)[name = tensor("attn_output_61_cast_fp16")]; tensor attn_output_63_perm_0 = const()[name = tensor("attn_output_63_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_971 = const()[name = tensor("op_971"), val = tensor([1, 77, 1024])]; tensor attn_output_63_cast_fp16 = transpose(perm = attn_output_63_perm_0, x = attn_output_61_cast_fp16)[name = tensor("transpose_167")]; tensor input_185_cast_fp16 = reshape(shape = var_971, x = attn_output_63_cast_fp16)[name = tensor("input_185_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(485561728)))]; tensor text_encoder_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487658944)))]; tensor linear_93_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16, x = input_185_cast_fp16)[name = tensor("linear_93_cast_fp16")]; tensor input_187_cast_fp16 = add(x = input_183_cast_fp16, y = linear_93_cast_fp16)[name = tensor("input_187_cast_fp16")]; tensor input_189_axes_0 = const()[name = tensor("input_189_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487661056)))]; tensor text_encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487663168)))]; tensor input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16, x = input_187_cast_fp16)[name = tensor("input_189_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487665280)))]; tensor text_encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496053952)))]; tensor linear_94_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16, x = input_189_cast_fp16)[name = tensor("linear_94_cast_fp16")]; tensor input_193_mode_0 = const()[name = tensor("input_193_mode_0"), val = tensor("EXACT")]; tensor input_193_cast_fp16 = gelu(mode = input_193_mode_0, x = linear_94_cast_fp16)[name = tensor("input_193_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496062208)))]; tensor text_encoder_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(504450880)))]; tensor linear_95_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16, x = input_193_cast_fp16)[name = tensor("linear_95_cast_fp16")]; tensor input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_95_cast_fp16)[name = tensor("input_195_cast_fp16")]; tensor hidden_states_97_axes_0 = const()[name = tensor("hidden_states_97_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(504452992)))]; tensor text_encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(504455104)))]; tensor hidden_states_97_cast_fp16 = layer_norm(axes = hidden_states_97_axes_0, beta = text_encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16, x = input_195_cast_fp16)[name = tensor("hidden_states_97_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(504457216)))]; tensor text_encoder_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506554432)))]; tensor linear_96_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor("linear_96_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506556544)))]; tensor text_encoder_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(508653760)))]; tensor linear_97_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor("linear_97_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(508655872)))]; tensor text_encoder_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510753088)))]; tensor linear_98_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor("linear_98_cast_fp16")]; tensor var_1015 = const()[name = tensor("op_1015"), val = tensor([1, -1, 16, 64])]; tensor var_1016_cast_fp16 = reshape(shape = var_1015, x = linear_96_cast_fp16)[name = tensor("op_1016_cast_fp16")]; tensor var_1018 = const()[name = tensor("op_1018"), val = tensor([1, -1, 16, 64])]; tensor var_1019_cast_fp16 = reshape(shape = var_1018, x = linear_97_cast_fp16)[name = tensor("op_1019_cast_fp16")]; tensor var_1021 = const()[name = tensor("op_1021"), val = tensor([1, -1, 16, 64])]; tensor var_1022_cast_fp16 = reshape(shape = var_1021, x = linear_98_cast_fp16)[name = tensor("op_1022_cast_fp16")]; tensor value_states_67_perm_0 = const()[name = tensor("value_states_67_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_16_cast_fp16 = mul(x = var_1016_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_16_cast_fp16")]; tensor matmul_16_transpose_y_0 = const()[name = tensor("matmul_16_transpose_y_0"), val = tensor(true)]; tensor matmul_16_transpose_x_0 = const()[name = tensor("matmul_16_transpose_x_0"), val = tensor(false)]; tensor transpose_125_perm_0 = const()[name = tensor("transpose_125_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_126_perm_0 = const()[name = tensor("transpose_126_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_126 = transpose(perm = transpose_126_perm_0, x = var_1019_cast_fp16)[name = tensor("transpose_164")]; tensor transpose_125 = transpose(perm = transpose_125_perm_0, x = mul_16_cast_fp16)[name = tensor("transpose_165")]; tensor matmul_16_cast_fp16 = matmul(transpose_x = matmul_16_transpose_x_0, transpose_y = matmul_16_transpose_y_0, x = transpose_125, y = transpose_126)[name = tensor("matmul_16_cast_fp16")]; tensor add_16_cast_fp16 = add(x = matmul_16_cast_fp16, y = var_58_to_fp16)[name = tensor("add_16_cast_fp16")]; tensor softmax_16_axis_0 = const()[name = tensor("softmax_16_axis_0"), val = tensor(-1)]; tensor softmax_16_cast_fp16 = softmax(axis = softmax_16_axis_0, x = add_16_cast_fp16)[name = tensor("softmax_16_cast_fp16")]; tensor attn_output_65_transpose_x_0 = const()[name = tensor("attn_output_65_transpose_x_0"), val = tensor(false)]; tensor attn_output_65_transpose_y_0 = const()[name = tensor("attn_output_65_transpose_y_0"), val = tensor(false)]; tensor value_states_67_cast_fp16 = transpose(perm = value_states_67_perm_0, x = var_1022_cast_fp16)[name = tensor("transpose_166")]; tensor attn_output_65_cast_fp16 = matmul(transpose_x = attn_output_65_transpose_x_0, transpose_y = attn_output_65_transpose_y_0, x = softmax_16_cast_fp16, y = value_states_67_cast_fp16)[name = tensor("attn_output_65_cast_fp16")]; tensor attn_output_67_perm_0 = const()[name = tensor("attn_output_67_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1026 = const()[name = tensor("op_1026"), val = tensor([1, 77, 1024])]; tensor attn_output_67_cast_fp16 = transpose(perm = attn_output_67_perm_0, x = attn_output_65_cast_fp16)[name = tensor("transpose_163")]; tensor input_197_cast_fp16 = reshape(shape = var_1026, x = attn_output_67_cast_fp16)[name = tensor("input_197_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510755200)))]; tensor text_encoder_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512852416)))]; tensor linear_99_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16, x = input_197_cast_fp16)[name = tensor("linear_99_cast_fp16")]; tensor input_199_cast_fp16 = add(x = input_195_cast_fp16, y = linear_99_cast_fp16)[name = tensor("input_199_cast_fp16")]; tensor input_201_axes_0 = const()[name = tensor("input_201_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512854528)))]; tensor text_encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512856640)))]; tensor input_201_cast_fp16 = layer_norm(axes = input_201_axes_0, beta = text_encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16, x = input_199_cast_fp16)[name = tensor("input_201_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512858752)))]; tensor text_encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521247424)))]; tensor linear_100_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16, x = input_201_cast_fp16)[name = tensor("linear_100_cast_fp16")]; tensor input_205_mode_0 = const()[name = tensor("input_205_mode_0"), val = tensor("EXACT")]; tensor input_205_cast_fp16 = gelu(mode = input_205_mode_0, x = linear_100_cast_fp16)[name = tensor("input_205_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521255680)))]; tensor text_encoder_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529644352)))]; tensor linear_101_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16, x = input_205_cast_fp16)[name = tensor("linear_101_cast_fp16")]; tensor input_207_cast_fp16 = add(x = input_199_cast_fp16, y = linear_101_cast_fp16)[name = tensor("input_207_cast_fp16")]; tensor hidden_states_103_axes_0 = const()[name = tensor("hidden_states_103_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529646464)))]; tensor text_encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529648576)))]; tensor hidden_states_103_cast_fp16 = layer_norm(axes = hidden_states_103_axes_0, beta = text_encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16, x = input_207_cast_fp16)[name = tensor("hidden_states_103_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529650688)))]; tensor text_encoder_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(531747904)))]; tensor linear_102_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor("linear_102_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(531750016)))]; tensor text_encoder_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533847232)))]; tensor linear_103_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor("linear_103_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533849344)))]; tensor text_encoder_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535946560)))]; tensor linear_104_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor("linear_104_cast_fp16")]; tensor var_1070 = const()[name = tensor("op_1070"), val = tensor([1, -1, 16, 64])]; tensor var_1071_cast_fp16 = reshape(shape = var_1070, x = linear_102_cast_fp16)[name = tensor("op_1071_cast_fp16")]; tensor var_1073 = const()[name = tensor("op_1073"), val = tensor([1, -1, 16, 64])]; tensor var_1074_cast_fp16 = reshape(shape = var_1073, x = linear_103_cast_fp16)[name = tensor("op_1074_cast_fp16")]; tensor var_1076 = const()[name = tensor("op_1076"), val = tensor([1, -1, 16, 64])]; tensor var_1077_cast_fp16 = reshape(shape = var_1076, x = linear_104_cast_fp16)[name = tensor("op_1077_cast_fp16")]; tensor value_states_71_perm_0 = const()[name = tensor("value_states_71_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_17_cast_fp16 = mul(x = var_1071_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_17_cast_fp16")]; tensor matmul_17_transpose_y_0 = const()[name = tensor("matmul_17_transpose_y_0"), val = tensor(true)]; tensor matmul_17_transpose_x_0 = const()[name = tensor("matmul_17_transpose_x_0"), val = tensor(false)]; tensor transpose_127_perm_0 = const()[name = tensor("transpose_127_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_128_perm_0 = const()[name = tensor("transpose_128_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_128 = transpose(perm = transpose_128_perm_0, x = var_1074_cast_fp16)[name = tensor("transpose_160")]; tensor transpose_127 = transpose(perm = transpose_127_perm_0, x = mul_17_cast_fp16)[name = tensor("transpose_161")]; tensor matmul_17_cast_fp16 = matmul(transpose_x = matmul_17_transpose_x_0, transpose_y = matmul_17_transpose_y_0, x = transpose_127, y = transpose_128)[name = tensor("matmul_17_cast_fp16")]; tensor add_17_cast_fp16 = add(x = matmul_17_cast_fp16, y = var_58_to_fp16)[name = tensor("add_17_cast_fp16")]; tensor softmax_17_axis_0 = const()[name = tensor("softmax_17_axis_0"), val = tensor(-1)]; tensor softmax_17_cast_fp16 = softmax(axis = softmax_17_axis_0, x = add_17_cast_fp16)[name = tensor("softmax_17_cast_fp16")]; tensor attn_output_69_transpose_x_0 = const()[name = tensor("attn_output_69_transpose_x_0"), val = tensor(false)]; tensor attn_output_69_transpose_y_0 = const()[name = tensor("attn_output_69_transpose_y_0"), val = tensor(false)]; tensor value_states_71_cast_fp16 = transpose(perm = value_states_71_perm_0, x = var_1077_cast_fp16)[name = tensor("transpose_162")]; tensor attn_output_69_cast_fp16 = matmul(transpose_x = attn_output_69_transpose_x_0, transpose_y = attn_output_69_transpose_y_0, x = softmax_17_cast_fp16, y = value_states_71_cast_fp16)[name = tensor("attn_output_69_cast_fp16")]; tensor attn_output_71_perm_0 = const()[name = tensor("attn_output_71_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1081 = const()[name = tensor("op_1081"), val = tensor([1, 77, 1024])]; tensor attn_output_71_cast_fp16 = transpose(perm = attn_output_71_perm_0, x = attn_output_69_cast_fp16)[name = tensor("transpose_159")]; tensor input_209_cast_fp16 = reshape(shape = var_1081, x = attn_output_71_cast_fp16)[name = tensor("input_209_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535948672)))]; tensor text_encoder_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538045888)))]; tensor linear_105_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16, x = input_209_cast_fp16)[name = tensor("linear_105_cast_fp16")]; tensor input_211_cast_fp16 = add(x = input_207_cast_fp16, y = linear_105_cast_fp16)[name = tensor("input_211_cast_fp16")]; tensor input_213_axes_0 = const()[name = tensor("input_213_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538048000)))]; tensor text_encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538050112)))]; tensor input_213_cast_fp16 = layer_norm(axes = input_213_axes_0, beta = text_encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16, x = input_211_cast_fp16)[name = tensor("input_213_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538052224)))]; tensor text_encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546440896)))]; tensor linear_106_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16, x = input_213_cast_fp16)[name = tensor("linear_106_cast_fp16")]; tensor input_217_mode_0 = const()[name = tensor("input_217_mode_0"), val = tensor("EXACT")]; tensor input_217_cast_fp16 = gelu(mode = input_217_mode_0, x = linear_106_cast_fp16)[name = tensor("input_217_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546449152)))]; tensor text_encoder_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554837824)))]; tensor linear_107_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16, x = input_217_cast_fp16)[name = tensor("linear_107_cast_fp16")]; tensor input_219_cast_fp16 = add(x = input_211_cast_fp16, y = linear_107_cast_fp16)[name = tensor("input_219_cast_fp16")]; tensor hidden_states_109_axes_0 = const()[name = tensor("hidden_states_109_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554839936)))]; tensor text_encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554842048)))]; tensor hidden_states_109_cast_fp16 = layer_norm(axes = hidden_states_109_axes_0, beta = text_encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16, x = input_219_cast_fp16)[name = tensor("hidden_states_109_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554844160)))]; tensor text_encoder_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(556941376)))]; tensor linear_108_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor("linear_108_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(556943488)))]; tensor text_encoder_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(559040704)))]; tensor linear_109_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor("linear_109_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(559042816)))]; tensor text_encoder_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(561140032)))]; tensor linear_110_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor("linear_110_cast_fp16")]; tensor var_1125 = const()[name = tensor("op_1125"), val = tensor([1, -1, 16, 64])]; tensor var_1126_cast_fp16 = reshape(shape = var_1125, x = linear_108_cast_fp16)[name = tensor("op_1126_cast_fp16")]; tensor var_1128 = const()[name = tensor("op_1128"), val = tensor([1, -1, 16, 64])]; tensor var_1129_cast_fp16 = reshape(shape = var_1128, x = linear_109_cast_fp16)[name = tensor("op_1129_cast_fp16")]; tensor var_1131 = const()[name = tensor("op_1131"), val = tensor([1, -1, 16, 64])]; tensor var_1132_cast_fp16 = reshape(shape = var_1131, x = linear_110_cast_fp16)[name = tensor("op_1132_cast_fp16")]; tensor value_states_75_perm_0 = const()[name = tensor("value_states_75_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_18_cast_fp16 = mul(x = var_1126_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_18_cast_fp16")]; tensor matmul_18_transpose_y_0 = const()[name = tensor("matmul_18_transpose_y_0"), val = tensor(true)]; tensor matmul_18_transpose_x_0 = const()[name = tensor("matmul_18_transpose_x_0"), val = tensor(false)]; tensor transpose_129_perm_0 = const()[name = tensor("transpose_129_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_130_perm_0 = const()[name = tensor("transpose_130_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_130 = transpose(perm = transpose_130_perm_0, x = var_1129_cast_fp16)[name = tensor("transpose_156")]; tensor transpose_129 = transpose(perm = transpose_129_perm_0, x = mul_18_cast_fp16)[name = tensor("transpose_157")]; tensor matmul_18_cast_fp16 = matmul(transpose_x = matmul_18_transpose_x_0, transpose_y = matmul_18_transpose_y_0, x = transpose_129, y = transpose_130)[name = tensor("matmul_18_cast_fp16")]; tensor add_18_cast_fp16 = add(x = matmul_18_cast_fp16, y = var_58_to_fp16)[name = tensor("add_18_cast_fp16")]; tensor softmax_18_axis_0 = const()[name = tensor("softmax_18_axis_0"), val = tensor(-1)]; tensor softmax_18_cast_fp16 = softmax(axis = softmax_18_axis_0, x = add_18_cast_fp16)[name = tensor("softmax_18_cast_fp16")]; tensor attn_output_73_transpose_x_0 = const()[name = tensor("attn_output_73_transpose_x_0"), val = tensor(false)]; tensor attn_output_73_transpose_y_0 = const()[name = tensor("attn_output_73_transpose_y_0"), val = tensor(false)]; tensor value_states_75_cast_fp16 = transpose(perm = value_states_75_perm_0, x = var_1132_cast_fp16)[name = tensor("transpose_158")]; tensor attn_output_73_cast_fp16 = matmul(transpose_x = attn_output_73_transpose_x_0, transpose_y = attn_output_73_transpose_y_0, x = softmax_18_cast_fp16, y = value_states_75_cast_fp16)[name = tensor("attn_output_73_cast_fp16")]; tensor attn_output_75_perm_0 = const()[name = tensor("attn_output_75_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1136 = const()[name = tensor("op_1136"), val = tensor([1, 77, 1024])]; tensor attn_output_75_cast_fp16 = transpose(perm = attn_output_75_perm_0, x = attn_output_73_cast_fp16)[name = tensor("transpose_155")]; tensor input_221_cast_fp16 = reshape(shape = var_1136, x = attn_output_75_cast_fp16)[name = tensor("input_221_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(561142144)))]; tensor text_encoder_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563239360)))]; tensor linear_111_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16, x = input_221_cast_fp16)[name = tensor("linear_111_cast_fp16")]; tensor input_223_cast_fp16 = add(x = input_219_cast_fp16, y = linear_111_cast_fp16)[name = tensor("input_223_cast_fp16")]; tensor input_225_axes_0 = const()[name = tensor("input_225_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563241472)))]; tensor text_encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563243584)))]; tensor input_225_cast_fp16 = layer_norm(axes = input_225_axes_0, beta = text_encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16, x = input_223_cast_fp16)[name = tensor("input_225_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563245696)))]; tensor text_encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571634368)))]; tensor linear_112_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16, x = input_225_cast_fp16)[name = tensor("linear_112_cast_fp16")]; tensor input_229_mode_0 = const()[name = tensor("input_229_mode_0"), val = tensor("EXACT")]; tensor input_229_cast_fp16 = gelu(mode = input_229_mode_0, x = linear_112_cast_fp16)[name = tensor("input_229_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571642624)))]; tensor text_encoder_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580031296)))]; tensor linear_113_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16, x = input_229_cast_fp16)[name = tensor("linear_113_cast_fp16")]; tensor input_231_cast_fp16 = add(x = input_223_cast_fp16, y = linear_113_cast_fp16)[name = tensor("input_231_cast_fp16")]; tensor hidden_states_115_axes_0 = const()[name = tensor("hidden_states_115_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580033408)))]; tensor text_encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580035520)))]; tensor hidden_states_115_cast_fp16 = layer_norm(axes = hidden_states_115_axes_0, beta = text_encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16, x = input_231_cast_fp16)[name = tensor("hidden_states_115_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580037632)))]; tensor text_encoder_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582134848)))]; tensor linear_114_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor("linear_114_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582136960)))]; tensor text_encoder_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(584234176)))]; tensor linear_115_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor("linear_115_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(584236288)))]; tensor text_encoder_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(586333504)))]; tensor linear_116_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor("linear_116_cast_fp16")]; tensor var_1180 = const()[name = tensor("op_1180"), val = tensor([1, -1, 16, 64])]; tensor var_1181_cast_fp16 = reshape(shape = var_1180, x = linear_114_cast_fp16)[name = tensor("op_1181_cast_fp16")]; tensor var_1183 = const()[name = tensor("op_1183"), val = tensor([1, -1, 16, 64])]; tensor var_1184_cast_fp16 = reshape(shape = var_1183, x = linear_115_cast_fp16)[name = tensor("op_1184_cast_fp16")]; tensor var_1186 = const()[name = tensor("op_1186"), val = tensor([1, -1, 16, 64])]; tensor var_1187_cast_fp16 = reshape(shape = var_1186, x = linear_116_cast_fp16)[name = tensor("op_1187_cast_fp16")]; tensor value_states_79_perm_0 = const()[name = tensor("value_states_79_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_19_cast_fp16 = mul(x = var_1181_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_19_cast_fp16")]; tensor matmul_19_transpose_y_0 = const()[name = tensor("matmul_19_transpose_y_0"), val = tensor(true)]; tensor matmul_19_transpose_x_0 = const()[name = tensor("matmul_19_transpose_x_0"), val = tensor(false)]; tensor transpose_131_perm_0 = const()[name = tensor("transpose_131_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_132_perm_0 = const()[name = tensor("transpose_132_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_132 = transpose(perm = transpose_132_perm_0, x = var_1184_cast_fp16)[name = tensor("transpose_152")]; tensor transpose_131 = transpose(perm = transpose_131_perm_0, x = mul_19_cast_fp16)[name = tensor("transpose_153")]; tensor matmul_19_cast_fp16 = matmul(transpose_x = matmul_19_transpose_x_0, transpose_y = matmul_19_transpose_y_0, x = transpose_131, y = transpose_132)[name = tensor("matmul_19_cast_fp16")]; tensor add_19_cast_fp16 = add(x = matmul_19_cast_fp16, y = var_58_to_fp16)[name = tensor("add_19_cast_fp16")]; tensor softmax_19_axis_0 = const()[name = tensor("softmax_19_axis_0"), val = tensor(-1)]; tensor softmax_19_cast_fp16 = softmax(axis = softmax_19_axis_0, x = add_19_cast_fp16)[name = tensor("softmax_19_cast_fp16")]; tensor attn_output_77_transpose_x_0 = const()[name = tensor("attn_output_77_transpose_x_0"), val = tensor(false)]; tensor attn_output_77_transpose_y_0 = const()[name = tensor("attn_output_77_transpose_y_0"), val = tensor(false)]; tensor value_states_79_cast_fp16 = transpose(perm = value_states_79_perm_0, x = var_1187_cast_fp16)[name = tensor("transpose_154")]; tensor attn_output_77_cast_fp16 = matmul(transpose_x = attn_output_77_transpose_x_0, transpose_y = attn_output_77_transpose_y_0, x = softmax_19_cast_fp16, y = value_states_79_cast_fp16)[name = tensor("attn_output_77_cast_fp16")]; tensor attn_output_79_perm_0 = const()[name = tensor("attn_output_79_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1191 = const()[name = tensor("op_1191"), val = tensor([1, 77, 1024])]; tensor attn_output_79_cast_fp16 = transpose(perm = attn_output_79_perm_0, x = attn_output_77_cast_fp16)[name = tensor("transpose_151")]; tensor input_233_cast_fp16 = reshape(shape = var_1191, x = attn_output_79_cast_fp16)[name = tensor("input_233_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(586335616)))]; tensor text_encoder_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(588432832)))]; tensor linear_117_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16, x = input_233_cast_fp16)[name = tensor("linear_117_cast_fp16")]; tensor input_235_cast_fp16 = add(x = input_231_cast_fp16, y = linear_117_cast_fp16)[name = tensor("input_235_cast_fp16")]; tensor input_237_axes_0 = const()[name = tensor("input_237_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(588434944)))]; tensor text_encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(588437056)))]; tensor input_237_cast_fp16 = layer_norm(axes = input_237_axes_0, beta = text_encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16, x = input_235_cast_fp16)[name = tensor("input_237_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(588439168)))]; tensor text_encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596827840)))]; tensor linear_118_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16, x = input_237_cast_fp16)[name = tensor("linear_118_cast_fp16")]; tensor input_241_mode_0 = const()[name = tensor("input_241_mode_0"), val = tensor("EXACT")]; tensor input_241_cast_fp16 = gelu(mode = input_241_mode_0, x = linear_118_cast_fp16)[name = tensor("input_241_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596836096)))]; tensor text_encoder_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(605224768)))]; tensor linear_119_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16, x = input_241_cast_fp16)[name = tensor("linear_119_cast_fp16")]; tensor input_243_cast_fp16 = add(x = input_235_cast_fp16, y = linear_119_cast_fp16)[name = tensor("input_243_cast_fp16")]; tensor hidden_states_121_axes_0 = const()[name = tensor("hidden_states_121_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(605226880)))]; tensor text_encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(605228992)))]; tensor hidden_states_121_cast_fp16 = layer_norm(axes = hidden_states_121_axes_0, beta = text_encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16, x = input_243_cast_fp16)[name = tensor("hidden_states_121_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(605231104)))]; tensor text_encoder_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607328320)))]; tensor linear_120_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor("linear_120_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607330432)))]; tensor text_encoder_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(609427648)))]; tensor linear_121_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor("linear_121_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(609429760)))]; tensor text_encoder_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611526976)))]; tensor linear_122_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor("linear_122_cast_fp16")]; tensor var_1235 = const()[name = tensor("op_1235"), val = tensor([1, -1, 16, 64])]; tensor var_1236_cast_fp16 = reshape(shape = var_1235, x = linear_120_cast_fp16)[name = tensor("op_1236_cast_fp16")]; tensor var_1238 = const()[name = tensor("op_1238"), val = tensor([1, -1, 16, 64])]; tensor var_1239_cast_fp16 = reshape(shape = var_1238, x = linear_121_cast_fp16)[name = tensor("op_1239_cast_fp16")]; tensor var_1241 = const()[name = tensor("op_1241"), val = tensor([1, -1, 16, 64])]; tensor var_1242_cast_fp16 = reshape(shape = var_1241, x = linear_122_cast_fp16)[name = tensor("op_1242_cast_fp16")]; tensor value_states_83_perm_0 = const()[name = tensor("value_states_83_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_20_cast_fp16 = mul(x = var_1236_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_20_cast_fp16")]; tensor matmul_20_transpose_y_0 = const()[name = tensor("matmul_20_transpose_y_0"), val = tensor(true)]; tensor matmul_20_transpose_x_0 = const()[name = tensor("matmul_20_transpose_x_0"), val = tensor(false)]; tensor transpose_133_perm_0 = const()[name = tensor("transpose_133_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_134_perm_0 = const()[name = tensor("transpose_134_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_134 = transpose(perm = transpose_134_perm_0, x = var_1239_cast_fp16)[name = tensor("transpose_148")]; tensor transpose_133 = transpose(perm = transpose_133_perm_0, x = mul_20_cast_fp16)[name = tensor("transpose_149")]; tensor matmul_20_cast_fp16 = matmul(transpose_x = matmul_20_transpose_x_0, transpose_y = matmul_20_transpose_y_0, x = transpose_133, y = transpose_134)[name = tensor("matmul_20_cast_fp16")]; tensor add_20_cast_fp16 = add(x = matmul_20_cast_fp16, y = var_58_to_fp16)[name = tensor("add_20_cast_fp16")]; tensor softmax_20_axis_0 = const()[name = tensor("softmax_20_axis_0"), val = tensor(-1)]; tensor softmax_20_cast_fp16 = softmax(axis = softmax_20_axis_0, x = add_20_cast_fp16)[name = tensor("softmax_20_cast_fp16")]; tensor attn_output_81_transpose_x_0 = const()[name = tensor("attn_output_81_transpose_x_0"), val = tensor(false)]; tensor attn_output_81_transpose_y_0 = const()[name = tensor("attn_output_81_transpose_y_0"), val = tensor(false)]; tensor value_states_83_cast_fp16 = transpose(perm = value_states_83_perm_0, x = var_1242_cast_fp16)[name = tensor("transpose_150")]; tensor attn_output_81_cast_fp16 = matmul(transpose_x = attn_output_81_transpose_x_0, transpose_y = attn_output_81_transpose_y_0, x = softmax_20_cast_fp16, y = value_states_83_cast_fp16)[name = tensor("attn_output_81_cast_fp16")]; tensor attn_output_83_perm_0 = const()[name = tensor("attn_output_83_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1246 = const()[name = tensor("op_1246"), val = tensor([1, 77, 1024])]; tensor attn_output_83_cast_fp16 = transpose(perm = attn_output_83_perm_0, x = attn_output_81_cast_fp16)[name = tensor("transpose_147")]; tensor input_245_cast_fp16 = reshape(shape = var_1246, x = attn_output_83_cast_fp16)[name = tensor("input_245_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611529088)))]; tensor text_encoder_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613626304)))]; tensor linear_123_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16, x = input_245_cast_fp16)[name = tensor("linear_123_cast_fp16")]; tensor input_247_cast_fp16 = add(x = input_243_cast_fp16, y = linear_123_cast_fp16)[name = tensor("input_247_cast_fp16")]; tensor input_249_axes_0 = const()[name = tensor("input_249_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613628416)))]; tensor text_encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613630528)))]; tensor input_249_cast_fp16 = layer_norm(axes = input_249_axes_0, beta = text_encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16, x = input_247_cast_fp16)[name = tensor("input_249_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613632640)))]; tensor text_encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622021312)))]; tensor linear_124_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16, x = input_249_cast_fp16)[name = tensor("linear_124_cast_fp16")]; tensor input_253_mode_0 = const()[name = tensor("input_253_mode_0"), val = tensor("EXACT")]; tensor input_253_cast_fp16 = gelu(mode = input_253_mode_0, x = linear_124_cast_fp16)[name = tensor("input_253_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622029568)))]; tensor text_encoder_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630418240)))]; tensor linear_125_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16, x = input_253_cast_fp16)[name = tensor("linear_125_cast_fp16")]; tensor input_255_cast_fp16 = add(x = input_247_cast_fp16, y = linear_125_cast_fp16)[name = tensor("input_255_cast_fp16")]; tensor hidden_states_127_axes_0 = const()[name = tensor("hidden_states_127_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630420352)))]; tensor text_encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630422464)))]; tensor hidden_states_127_cast_fp16 = layer_norm(axes = hidden_states_127_axes_0, beta = text_encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16, x = input_255_cast_fp16)[name = tensor("hidden_states_127_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630424576)))]; tensor text_encoder_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632521792)))]; tensor linear_126_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor("linear_126_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632523904)))]; tensor text_encoder_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(634621120)))]; tensor linear_127_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor("linear_127_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(634623232)))]; tensor text_encoder_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(636720448)))]; tensor linear_128_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor("linear_128_cast_fp16")]; tensor var_1290 = const()[name = tensor("op_1290"), val = tensor([1, -1, 16, 64])]; tensor var_1291_cast_fp16 = reshape(shape = var_1290, x = linear_126_cast_fp16)[name = tensor("op_1291_cast_fp16")]; tensor var_1293 = const()[name = tensor("op_1293"), val = tensor([1, -1, 16, 64])]; tensor var_1294_cast_fp16 = reshape(shape = var_1293, x = linear_127_cast_fp16)[name = tensor("op_1294_cast_fp16")]; tensor var_1296 = const()[name = tensor("op_1296"), val = tensor([1, -1, 16, 64])]; tensor var_1297_cast_fp16 = reshape(shape = var_1296, x = linear_128_cast_fp16)[name = tensor("op_1297_cast_fp16")]; tensor value_states_87_perm_0 = const()[name = tensor("value_states_87_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_21_cast_fp16 = mul(x = var_1291_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_21_cast_fp16")]; tensor matmul_21_transpose_y_0 = const()[name = tensor("matmul_21_transpose_y_0"), val = tensor(true)]; tensor matmul_21_transpose_x_0 = const()[name = tensor("matmul_21_transpose_x_0"), val = tensor(false)]; tensor transpose_135_perm_0 = const()[name = tensor("transpose_135_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_136_perm_0 = const()[name = tensor("transpose_136_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_136 = transpose(perm = transpose_136_perm_0, x = var_1294_cast_fp16)[name = tensor("transpose_144")]; tensor transpose_135 = transpose(perm = transpose_135_perm_0, x = mul_21_cast_fp16)[name = tensor("transpose_145")]; tensor matmul_21_cast_fp16 = matmul(transpose_x = matmul_21_transpose_x_0, transpose_y = matmul_21_transpose_y_0, x = transpose_135, y = transpose_136)[name = tensor("matmul_21_cast_fp16")]; tensor add_21_cast_fp16 = add(x = matmul_21_cast_fp16, y = var_58_to_fp16)[name = tensor("add_21_cast_fp16")]; tensor softmax_21_axis_0 = const()[name = tensor("softmax_21_axis_0"), val = tensor(-1)]; tensor softmax_21_cast_fp16 = softmax(axis = softmax_21_axis_0, x = add_21_cast_fp16)[name = tensor("softmax_21_cast_fp16")]; tensor attn_output_85_transpose_x_0 = const()[name = tensor("attn_output_85_transpose_x_0"), val = tensor(false)]; tensor attn_output_85_transpose_y_0 = const()[name = tensor("attn_output_85_transpose_y_0"), val = tensor(false)]; tensor value_states_87_cast_fp16 = transpose(perm = value_states_87_perm_0, x = var_1297_cast_fp16)[name = tensor("transpose_146")]; tensor attn_output_85_cast_fp16 = matmul(transpose_x = attn_output_85_transpose_x_0, transpose_y = attn_output_85_transpose_y_0, x = softmax_21_cast_fp16, y = value_states_87_cast_fp16)[name = tensor("attn_output_85_cast_fp16")]; tensor attn_output_87_perm_0 = const()[name = tensor("attn_output_87_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1301 = const()[name = tensor("op_1301"), val = tensor([1, 77, 1024])]; tensor attn_output_87_cast_fp16 = transpose(perm = attn_output_87_perm_0, x = attn_output_85_cast_fp16)[name = tensor("transpose_143")]; tensor input_257_cast_fp16 = reshape(shape = var_1301, x = attn_output_87_cast_fp16)[name = tensor("input_257_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(636722560)))]; tensor text_encoder_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638819776)))]; tensor linear_129_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16, x = input_257_cast_fp16)[name = tensor("linear_129_cast_fp16")]; tensor input_259_cast_fp16 = add(x = input_255_cast_fp16, y = linear_129_cast_fp16)[name = tensor("input_259_cast_fp16")]; tensor input_261_axes_0 = const()[name = tensor("input_261_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638821888)))]; tensor text_encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638824000)))]; tensor input_261_cast_fp16 = layer_norm(axes = input_261_axes_0, beta = text_encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16, x = input_259_cast_fp16)[name = tensor("input_261_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638826112)))]; tensor text_encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647214784)))]; tensor linear_130_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16, x = input_261_cast_fp16)[name = tensor("linear_130_cast_fp16")]; tensor input_265_mode_0 = const()[name = tensor("input_265_mode_0"), val = tensor("EXACT")]; tensor input_265_cast_fp16 = gelu(mode = input_265_mode_0, x = linear_130_cast_fp16)[name = tensor("input_265_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647223040)))]; tensor text_encoder_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(655611712)))]; tensor linear_131_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16, x = input_265_cast_fp16)[name = tensor("linear_131_cast_fp16")]; tensor input_267_cast_fp16 = add(x = input_259_cast_fp16, y = linear_131_cast_fp16)[name = tensor("input_267_cast_fp16")]; tensor hidden_states_133_axes_0 = const()[name = tensor("hidden_states_133_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(655613824)))]; tensor text_encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(655615936)))]; tensor hidden_states_133_cast_fp16 = layer_norm(axes = hidden_states_133_axes_0, beta = text_encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16, x = input_267_cast_fp16)[name = tensor("hidden_states_133_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(655618048)))]; tensor text_encoder_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657715264)))]; tensor linear_132_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16, x = hidden_states_133_cast_fp16)[name = tensor("linear_132_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657717376)))]; tensor text_encoder_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(659814592)))]; tensor linear_133_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16, x = hidden_states_133_cast_fp16)[name = tensor("linear_133_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(659816704)))]; tensor text_encoder_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(661913920)))]; tensor linear_134_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16, x = hidden_states_133_cast_fp16)[name = tensor("linear_134_cast_fp16")]; tensor var_1345 = const()[name = tensor("op_1345"), val = tensor([1, -1, 16, 64])]; tensor var_1346_cast_fp16 = reshape(shape = var_1345, x = linear_132_cast_fp16)[name = tensor("op_1346_cast_fp16")]; tensor var_1348 = const()[name = tensor("op_1348"), val = tensor([1, -1, 16, 64])]; tensor var_1349_cast_fp16 = reshape(shape = var_1348, x = linear_133_cast_fp16)[name = tensor("op_1349_cast_fp16")]; tensor var_1351 = const()[name = tensor("op_1351"), val = tensor([1, -1, 16, 64])]; tensor var_1352_cast_fp16 = reshape(shape = var_1351, x = linear_134_cast_fp16)[name = tensor("op_1352_cast_fp16")]; tensor value_states_perm_0 = const()[name = tensor("value_states_perm_0"), val = tensor([0, 2, 1, 3])]; tensor mul_22_cast_fp16 = mul(x = var_1346_cast_fp16, y = var_17_to_fp16)[name = tensor("mul_22_cast_fp16")]; tensor matmul_22_transpose_y_0 = const()[name = tensor("matmul_22_transpose_y_0"), val = tensor(true)]; tensor matmul_22_transpose_x_0 = const()[name = tensor("matmul_22_transpose_x_0"), val = tensor(false)]; tensor transpose_137_perm_0 = const()[name = tensor("transpose_137_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_138_perm_0 = const()[name = tensor("transpose_138_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_138 = transpose(perm = transpose_138_perm_0, x = var_1349_cast_fp16)[name = tensor("transpose_140")]; tensor transpose_137 = transpose(perm = transpose_137_perm_0, x = mul_22_cast_fp16)[name = tensor("transpose_141")]; tensor matmul_22_cast_fp16 = matmul(transpose_x = matmul_22_transpose_x_0, transpose_y = matmul_22_transpose_y_0, x = transpose_137, y = transpose_138)[name = tensor("matmul_22_cast_fp16")]; tensor add_22_cast_fp16 = add(x = matmul_22_cast_fp16, y = var_58_to_fp16)[name = tensor("add_22_cast_fp16")]; tensor softmax_22_axis_0 = const()[name = tensor("softmax_22_axis_0"), val = tensor(-1)]; tensor softmax_22_cast_fp16 = softmax(axis = softmax_22_axis_0, x = add_22_cast_fp16)[name = tensor("softmax_22_cast_fp16")]; tensor attn_output_89_transpose_x_0 = const()[name = tensor("attn_output_89_transpose_x_0"), val = tensor(false)]; tensor attn_output_89_transpose_y_0 = const()[name = tensor("attn_output_89_transpose_y_0"), val = tensor(false)]; tensor value_states_cast_fp16 = transpose(perm = value_states_perm_0, x = var_1352_cast_fp16)[name = tensor("transpose_142")]; tensor attn_output_89_cast_fp16 = matmul(transpose_x = attn_output_89_transpose_x_0, transpose_y = attn_output_89_transpose_y_0, x = softmax_22_cast_fp16, y = value_states_cast_fp16)[name = tensor("attn_output_89_cast_fp16")]; tensor attn_output_perm_0 = const()[name = tensor("attn_output_perm_0"), val = tensor([0, 2, 1, 3])]; tensor var_1356 = const()[name = tensor("op_1356"), val = tensor([1, 77, 1024])]; tensor attn_output_cast_fp16 = transpose(perm = attn_output_perm_0, x = attn_output_89_cast_fp16)[name = tensor("transpose_139")]; tensor input_269_cast_fp16 = reshape(shape = var_1356, x = attn_output_cast_fp16)[name = tensor("input_269_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(661916032)))]; tensor text_encoder_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(664013248)))]; tensor linear_135_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16, x = input_269_cast_fp16)[name = tensor("linear_135_cast_fp16")]; tensor input_271_cast_fp16 = add(x = input_267_cast_fp16, y = linear_135_cast_fp16)[name = tensor("input_271_cast_fp16")]; tensor input_273_axes_0 = const()[name = tensor("input_273_axes_0"), val = tensor([-1])]; tensor text_encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(664015360)))]; tensor text_encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(664017472)))]; tensor input_273_cast_fp16 = layer_norm(axes = input_273_axes_0, beta = text_encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16, x = input_271_cast_fp16)[name = tensor("input_273_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(664019584)))]; tensor text_encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(672408256)))]; tensor linear_136_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16, x = input_273_cast_fp16)[name = tensor("linear_136_cast_fp16")]; tensor input_277_mode_0 = const()[name = tensor("input_277_mode_0"), val = tensor("EXACT")]; tensor input_277_cast_fp16 = gelu(mode = input_277_mode_0, x = linear_136_cast_fp16)[name = tensor("input_277_cast_fp16")]; tensor text_encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(672416512)))]; tensor text_encoder_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(680805184)))]; tensor linear_137_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16, x = input_277_cast_fp16)[name = tensor("linear_137_cast_fp16")]; tensor input_cast_fp16 = add(x = input_271_cast_fp16, y = linear_137_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(680807296)))]; 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(680809408)))]; 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_1381 = const()[name = tensor("op_1381"), val = tensor([0])]; tensor var_1383_axis_0 = const()[name = tensor("op_1383_axis_0"), val = tensor(-1)]; tensor var_1383_keep_dims_0 = const()[name = tensor("op_1383_keep_dims_0"), val = tensor(false)]; tensor var_1383 = reduce_argmax(axis = var_1383_axis_0, keep_dims = var_1383_keep_dims_0, x = cast_1)[name = tensor("op_1383")]; 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_1381, var_1383))[name = tensor("stack_0")]; tensor var_1385_transpose_batch_dims_0 = const()[name = tensor("op_1385_transpose_batch_dims_0"), val = tensor(0)]; tensor var_1385_transpose_cast_fp16 = gather_nd(batch_dims = var_1385_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast_fp16)[name = tensor("op_1385_transpose_cast_fp16")]; tensor var_1385_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_1385_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; tensor pooled_outputs = cast(dtype = var_1385_cast_fp16_to_fp32_dtype_0, x = var_1385_transpose_cast_fp16)[name = tensor("cast_98")]; tensor last_hidden_state = cast(dtype = last_hidden_state_cast_fp16_to_fp32_dtype_0, x = last_hidden_state_cast_fp16)[name = tensor("cast_99")]; } -> (last_hidden_state, pooled_outputs); }