program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.5.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})] { func main(tensor cache_length, tensor decoder_key_padding_mask, tensor encoder_output_embeds, tensor input_ids, tensor key_cache, tensor kv_cache_update_mask, tensor value_cache) { tensor var_80_axis_0 = const()[name = tensor("op_80_axis_0"), val = tensor(0)]; tensor var_80_batch_dims_0 = const()[name = tensor("op_80_batch_dims_0"), val = tensor(0)]; tensor embed_tokens_weight_to_fp16 = const()[name = tensor("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; tensor var_80_cast_fp16 = gather(axis = var_80_axis_0, batch_dims = var_80_batch_dims_0, indices = input_ids, x = embed_tokens_weight_to_fp16)[name = tensor("op_80_cast_fp16")]; tensor var_84_axis_0 = const()[name = tensor("op_84_axis_0"), val = tensor(0)]; tensor var_84_batch_dims_0 = const()[name = tensor("op_84_batch_dims_0"), val = tensor(0)]; tensor embed_positions_weight_to_fp16 = const()[name = tensor("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132774528)))]; tensor var_84_cast_fp16 = gather(axis = var_84_axis_0, batch_dims = var_84_batch_dims_0, indices = cache_length, x = embed_positions_weight_to_fp16)[name = tensor("op_84_cast_fp16")]; tensor hidden_states_1_cast_fp16 = add(x = var_80_cast_fp16, y = var_84_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; tensor var_98_axes_0 = const()[name = tensor("op_98_axes_0"), val = tensor([2])]; tensor var_98_cast_fp16 = expand_dims(axes = var_98_axes_0, x = hidden_states_1_cast_fp16)[name = tensor("op_98_cast_fp16")]; tensor inputs_1_axes_0 = const()[name = tensor("inputs_1_axes_0"), val = tensor([3])]; tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_98_cast_fp16)[name = tensor("inputs_1_cast_fp16")]; tensor tile_0 = const()[name = tensor("tile_0"), val = tensor([1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280])]; tensor var_103_axis_0 = const()[name = tensor("op_103_axis_0"), val = tensor(1)]; tensor var_103_cast_fp16_0, tensor var_103_cast_fp16_1, tensor var_103_cast_fp16_2, tensor var_103_cast_fp16_3, tensor var_103_cast_fp16_4, tensor var_103_cast_fp16_5, tensor var_103_cast_fp16_6, tensor var_103_cast_fp16_7, tensor var_103_cast_fp16_8, tensor var_103_cast_fp16_9, tensor var_103_cast_fp16_10, tensor var_103_cast_fp16_11, tensor var_103_cast_fp16_12, tensor var_103_cast_fp16_13, tensor var_103_cast_fp16_14, tensor var_103_cast_fp16_15, tensor var_103_cast_fp16_16, tensor var_103_cast_fp16_17, tensor var_103_cast_fp16_18, tensor var_103_cast_fp16_19, tensor var_103_cast_fp16_20, tensor var_103_cast_fp16_21, tensor var_103_cast_fp16_22, tensor var_103_cast_fp16_23, tensor var_103_cast_fp16_24, tensor var_103_cast_fp16_25, tensor var_103_cast_fp16_26, tensor var_103_cast_fp16_27, tensor var_103_cast_fp16_28, tensor var_103_cast_fp16_29, tensor var_103_cast_fp16_30, tensor var_103_cast_fp16_31 = split(axis = var_103_axis_0, split_sizes = tile_0, x = key_cache)[name = tensor("op_103_cast_fp16")]; tensor tile_1 = const()[name = tensor("tile_1"), val = tensor([1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280, 1280])]; tensor var_138_axis_0 = const()[name = tensor("op_138_axis_0"), val = tensor(1)]; tensor var_138_cast_fp16_0, tensor var_138_cast_fp16_1, tensor var_138_cast_fp16_2, tensor var_138_cast_fp16_3, tensor var_138_cast_fp16_4, tensor var_138_cast_fp16_5, tensor var_138_cast_fp16_6, tensor var_138_cast_fp16_7, tensor var_138_cast_fp16_8, tensor var_138_cast_fp16_9, tensor var_138_cast_fp16_10, tensor var_138_cast_fp16_11, tensor var_138_cast_fp16_12, tensor var_138_cast_fp16_13, tensor var_138_cast_fp16_14, tensor var_138_cast_fp16_15, tensor var_138_cast_fp16_16, tensor var_138_cast_fp16_17, tensor var_138_cast_fp16_18, tensor var_138_cast_fp16_19, tensor var_138_cast_fp16_20, tensor var_138_cast_fp16_21, tensor var_138_cast_fp16_22, tensor var_138_cast_fp16_23, tensor var_138_cast_fp16_24, tensor var_138_cast_fp16_25, tensor var_138_cast_fp16_26, tensor var_138_cast_fp16_27, tensor var_138_cast_fp16_28, tensor var_138_cast_fp16_29, tensor var_138_cast_fp16_30, tensor var_138_cast_fp16_31 = split(axis = var_138_axis_0, split_sizes = tile_1, x = value_cache)[name = tensor("op_138_cast_fp16")]; tensor var_176 = const()[name = tensor("op_176"), val = tensor(3)]; tensor out_1_axes_0 = const()[name = tensor("out_1_axes_0"), val = tensor([1])]; tensor var_201_to_fp16 = const()[name = tensor("op_201_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_201_to_fp16, x = inputs_1_cast_fp16)[name = tensor("out_1_cast_fp16")]; tensor obj_1_mean_0_to_fp16 = const()[name = tensor("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133921472)))]; tensor obj_1_variance_0_to_fp16 = const()[name = tensor("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133924096)))]; tensor obj_1_gamma_0_to_fp16 = const()[name = tensor("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133926720)))]; tensor obj_1_beta_0_to_fp16 = const()[name = tensor("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133929344)))]; tensor obj_1_epsilon_0_to_fp16 = const()[name = tensor("obj_1_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = tensor("obj_1_cast_fp16")]; tensor query_1_pad_type_0 = const()[name = tensor("query_1_pad_type_0"), val = tensor("valid")]; tensor query_1_strides_0 = const()[name = tensor("query_1_strides_0"), val = tensor([1, 1])]; tensor query_1_pad_0 = const()[name = tensor("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_1_dilations_0 = const()[name = tensor("query_1_dilations_0"), val = tensor([1, 1])]; tensor query_1_groups_0 = const()[name = tensor("query_1_groups_0"), val = tensor(1)]; tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133931968)))]; tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137208832)))]; tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor("query_1_cast_fp16")]; tensor current_key_1_pad_type_0 = const()[name = tensor("current_key_1_pad_type_0"), val = tensor("valid")]; tensor current_key_1_strides_0 = const()[name = tensor("current_key_1_strides_0"), val = tensor([1, 1])]; tensor current_key_1_pad_0 = const()[name = tensor("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_1_dilations_0 = const()[name = tensor("current_key_1_dilations_0"), val = tensor([1, 1])]; tensor current_key_1_groups_0 = const()[name = tensor("current_key_1_groups_0"), val = tensor(1)]; tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137211456)))]; tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor("current_key_1_cast_fp16")]; tensor current_value_1_pad_type_0 = const()[name = tensor("current_value_1_pad_type_0"), val = tensor("valid")]; tensor current_value_1_strides_0 = const()[name = tensor("current_value_1_strides_0"), val = tensor([1, 1])]; tensor current_value_1_pad_0 = const()[name = tensor("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_1_dilations_0 = const()[name = tensor("current_value_1_dilations_0"), val = tensor([1, 1])]; tensor current_value_1_groups_0 = const()[name = tensor("current_value_1_groups_0"), val = tensor(1)]; tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140488320)))]; tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143765184)))]; tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor("current_value_1_cast_fp16")]; tensor var_236_axes_0 = const()[name = tensor("op_236_axes_0"), val = tensor([1])]; tensor var_236_cast_fp16 = expand_dims(axes = var_236_axes_0, x = kv_cache_update_mask)[name = tensor("op_236_cast_fp16")]; tensor var_237_axes_0 = const()[name = tensor("op_237_axes_0"), val = tensor([2])]; tensor var_237_cast_fp16 = expand_dims(axes = var_237_axes_0, x = var_236_cast_fp16)[name = tensor("op_237_cast_fp16")]; tensor var_177_to_fp16 = const()[name = tensor("op_177_to_fp16"), val = tensor(0x1p+0)]; tensor var_239_cast_fp16 = sub(x = var_177_to_fp16, y = var_237_cast_fp16)[name = tensor("op_239_cast_fp16")]; tensor var_240_cast_fp16 = mul(x = var_103_cast_fp16_0, y = var_239_cast_fp16)[name = tensor("op_240_cast_fp16")]; tensor var_241_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_241_cast_fp16")]; tensor key_1_cast_fp16 = add(x = var_240_cast_fp16, y = var_241_cast_fp16)[name = tensor("key_1_cast_fp16")]; tensor var_244_cast_fp16 = mul(x = var_138_cast_fp16_0, y = var_239_cast_fp16)[name = tensor("op_244_cast_fp16")]; tensor var_245_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_245_cast_fp16")]; tensor value_1_cast_fp16 = add(x = var_244_cast_fp16, y = var_245_cast_fp16)[name = tensor("value_1_cast_fp16")]; tensor var_249 = const()[name = tensor("op_249"), val = tensor([1, 20, 64, 1])]; tensor mh_q_1_cast_fp16 = reshape(shape = var_249, x = query_1_cast_fp16)[name = tensor("mh_q_1_cast_fp16")]; tensor var_251_to_fp16 = const()[name = tensor("op_251_to_fp16"), val = tensor(0x1p-3)]; tensor var_252_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_251_to_fp16)[name = tensor("op_252_cast_fp16")]; tensor var_255 = const()[name = tensor("op_255"), val = tensor([1, 20, 64, 448])]; tensor var_256_cast_fp16 = reshape(shape = var_255, x = key_1_cast_fp16)[name = tensor("op_256_cast_fp16")]; tensor mh_w_1_transpose_x_0 = const()[name = tensor("mh_w_1_transpose_x_0"), val = tensor(true)]; tensor mh_w_1_transpose_y_0 = const()[name = tensor("mh_w_1_transpose_y_0"), val = tensor(false)]; tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_252_cast_fp16, y = var_256_cast_fp16)[name = tensor("mh_w_1_cast_fp16")]; tensor var_260_axes_0 = const()[name = tensor("op_260_axes_0"), val = tensor([1])]; tensor var_260_cast_fp16 = expand_dims(axes = var_260_axes_0, x = decoder_key_padding_mask)[name = tensor("op_260_cast_fp16")]; tensor var_261_axes_0 = const()[name = tensor("op_261_axes_0"), val = tensor([2])]; tensor var_261_cast_fp16 = expand_dims(axes = var_261_axes_0, x = var_260_cast_fp16)[name = tensor("op_261_cast_fp16")]; tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_3_cast_fp16")]; tensor var_264_cast_fp16 = softmax(axis = var_176, x = mh_w_3_cast_fp16)[name = tensor("op_264_cast_fp16")]; tensor var_265 = const()[name = tensor("op_265"), val = tensor([1, 20, 64, 448])]; tensor var_266_cast_fp16 = reshape(shape = var_265, x = value_1_cast_fp16)[name = tensor("op_266_cast_fp16")]; tensor attn_1_transpose_x_0 = const()[name = tensor("attn_1_transpose_x_0"), val = tensor(false)]; tensor attn_1_transpose_y_0 = const()[name = tensor("attn_1_transpose_y_0"), val = tensor(true)]; tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_266_cast_fp16, y = var_264_cast_fp16)[name = tensor("attn_1_cast_fp16")]; tensor var_269 = const()[name = tensor("op_269"), val = tensor([1, 1280, 1, 1])]; tensor input_1_cast_fp16 = reshape(shape = var_269, x = attn_1_cast_fp16)[name = tensor("input_1_cast_fp16")]; tensor obj_7_pad_type_0 = const()[name = tensor("obj_7_pad_type_0"), val = tensor("valid")]; tensor obj_7_strides_0 = const()[name = tensor("obj_7_strides_0"), val = tensor([1, 1])]; tensor obj_7_pad_0 = const()[name = tensor("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_7_dilations_0 = const()[name = tensor("obj_7_dilations_0"), val = tensor([1, 1])]; tensor obj_7_groups_0 = const()[name = tensor("obj_7_groups_0"), val = tensor(1)]; tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143767808)))]; tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147044672)))]; tensor obj_7_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = tensor("obj_7_cast_fp16")]; tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_7_cast_fp16)[name = tensor("inputs_3_cast_fp16")]; tensor out_3_axes_0 = const()[name = tensor("out_3_axes_0"), val = tensor([1])]; tensor var_291_to_fp16 = const()[name = tensor("op_291_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_291_to_fp16, x = inputs_3_cast_fp16)[name = tensor("out_3_cast_fp16")]; tensor obj_9_gamma_0_to_fp16 = const()[name = tensor("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147047296)))]; tensor obj_9_beta_0_to_fp16 = const()[name = tensor("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147049920)))]; tensor obj_9_epsilon_0_to_fp16 = const()[name = tensor("obj_9_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = tensor("obj_9_cast_fp16")]; tensor query_3_pad_type_0 = const()[name = tensor("query_3_pad_type_0"), val = tensor("valid")]; tensor query_3_strides_0 = const()[name = tensor("query_3_strides_0"), val = tensor([1, 1])]; tensor query_3_pad_0 = const()[name = tensor("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_3_dilations_0 = const()[name = tensor("query_3_dilations_0"), val = tensor([1, 1])]; tensor query_3_groups_0 = const()[name = tensor("query_3_groups_0"), val = tensor(1)]; tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147052544)))]; tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150329408)))]; tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor("query_3_cast_fp16")]; tensor key_3_pad_type_0 = const()[name = tensor("key_3_pad_type_0"), val = tensor("valid")]; tensor key_3_strides_0 = const()[name = tensor("key_3_strides_0"), val = tensor([1, 1])]; tensor key_3_pad_0 = const()[name = tensor("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_3_dilations_0 = const()[name = tensor("key_3_dilations_0"), val = tensor([1, 1])]; tensor key_3_groups_0 = const()[name = tensor("key_3_groups_0"), val = tensor(1)]; tensor layers_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150332032)))]; tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_3_cast_fp16")]; tensor value_3_pad_type_0 = const()[name = tensor("value_3_pad_type_0"), val = tensor("valid")]; tensor value_3_strides_0 = const()[name = tensor("value_3_strides_0"), val = tensor([1, 1])]; tensor value_3_pad_0 = const()[name = tensor("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_3_dilations_0 = const()[name = tensor("value_3_dilations_0"), val = tensor([1, 1])]; tensor value_3_groups_0 = const()[name = tensor("value_3_groups_0"), val = tensor(1)]; tensor layers_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153608896)))]; tensor layers_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156885760)))]; tensor value_3_cast_fp16 = conv(bias = layers_0_encoder_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_3_cast_fp16")]; tensor var_327 = const()[name = tensor("op_327"), val = tensor([1, 20, 64, 1])]; tensor mh_q_3_cast_fp16 = reshape(shape = var_327, x = query_3_cast_fp16)[name = tensor("mh_q_3_cast_fp16")]; tensor var_329_to_fp16 = const()[name = tensor("op_329_to_fp16"), val = tensor(0x1p-3)]; tensor var_330_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_329_to_fp16)[name = tensor("op_330_cast_fp16")]; tensor var_333 = const()[name = tensor("op_333"), val = tensor([1, 20, 64, 1500])]; tensor var_334_cast_fp16 = reshape(shape = var_333, x = key_3_cast_fp16)[name = tensor("op_334_cast_fp16")]; tensor mh_w_5_transpose_x_0 = const()[name = tensor("mh_w_5_transpose_x_0"), val = tensor(true)]; tensor mh_w_5_transpose_y_0 = const()[name = tensor("mh_w_5_transpose_y_0"), val = tensor(false)]; tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_330_cast_fp16, y = var_334_cast_fp16)[name = tensor("mh_w_5_cast_fp16")]; tensor obj_13_cast_fp16 = softmax(axis = var_176, x = mh_w_5_cast_fp16)[name = tensor("obj_13_cast_fp16")]; tensor var_338 = const()[name = tensor("op_338"), val = tensor([1, 20, 64, 1500])]; tensor var_339_cast_fp16 = reshape(shape = var_338, x = value_3_cast_fp16)[name = tensor("op_339_cast_fp16")]; tensor attn_3_transpose_x_0 = const()[name = tensor("attn_3_transpose_x_0"), val = tensor(false)]; tensor attn_3_transpose_y_0 = const()[name = tensor("attn_3_transpose_y_0"), val = tensor(true)]; tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_339_cast_fp16, y = obj_13_cast_fp16)[name = tensor("attn_3_cast_fp16")]; tensor var_342 = const()[name = tensor("op_342"), val = tensor([1, 1280, 1, 1])]; tensor input_3_cast_fp16 = reshape(shape = var_342, x = attn_3_cast_fp16)[name = tensor("input_3_cast_fp16")]; tensor obj_11_pad_type_0 = const()[name = tensor("obj_11_pad_type_0"), val = tensor("valid")]; tensor obj_11_strides_0 = const()[name = tensor("obj_11_strides_0"), val = tensor([1, 1])]; tensor obj_11_pad_0 = const()[name = tensor("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_11_dilations_0 = const()[name = tensor("obj_11_dilations_0"), val = tensor([1, 1])]; tensor obj_11_groups_0 = const()[name = tensor("obj_11_groups_0"), val = tensor(1)]; tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156888384)))]; tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160165248)))]; tensor obj_11_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = tensor("obj_11_cast_fp16")]; tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_11_cast_fp16)[name = tensor("inputs_5_cast_fp16")]; tensor out_5_axes_0 = const()[name = tensor("out_5_axes_0"), val = tensor([1])]; tensor var_360_to_fp16 = const()[name = tensor("op_360_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_360_to_fp16, x = inputs_5_cast_fp16)[name = tensor("out_5_cast_fp16")]; tensor input_5_gamma_0_to_fp16 = const()[name = tensor("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160167872)))]; tensor input_5_beta_0_to_fp16 = const()[name = tensor("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160170496)))]; tensor input_5_epsilon_0_to_fp16 = const()[name = tensor("input_5_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = tensor("input_5_cast_fp16")]; tensor input_7_pad_type_0 = const()[name = tensor("input_7_pad_type_0"), val = tensor("valid")]; tensor input_7_strides_0 = const()[name = tensor("input_7_strides_0"), val = tensor([1, 1])]; tensor input_7_pad_0 = const()[name = tensor("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_7_dilations_0 = const()[name = tensor("input_7_dilations_0"), val = tensor([1, 1])]; tensor input_7_groups_0 = const()[name = tensor("input_7_groups_0"), val = tensor(1)]; tensor layers_0_fc1_weight_to_fp16 = const()[name = tensor("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(160173120)))]; tensor layers_0_fc1_bias_to_fp16 = const()[name = tensor("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173280384)))]; tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = tensor("input_7_cast_fp16")]; tensor input_9_mode_0 = const()[name = tensor("input_9_mode_0"), val = tensor("EXACT")]; tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = tensor("input_9_cast_fp16")]; tensor hidden_states_3_pad_type_0 = const()[name = tensor("hidden_states_3_pad_type_0"), val = tensor("valid")]; tensor hidden_states_3_strides_0 = const()[name = tensor("hidden_states_3_strides_0"), val = tensor([1, 1])]; tensor hidden_states_3_pad_0 = const()[name = tensor("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_3_dilations_0 = const()[name = tensor("hidden_states_3_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_3_groups_0 = const()[name = tensor("hidden_states_3_groups_0"), val = tensor(1)]; tensor layers_0_fc2_weight_to_fp16 = const()[name = tensor("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173290688)))]; tensor layers_0_fc2_bias_to_fp16 = const()[name = tensor("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186397952)))]; tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = tensor("hidden_states_3_cast_fp16")]; tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = tensor("inputs_7_cast_fp16")]; tensor var_395 = const()[name = tensor("op_395"), val = tensor(3)]; tensor out_7_axes_0 = const()[name = tensor("out_7_axes_0"), val = tensor([1])]; tensor var_420_to_fp16 = const()[name = tensor("op_420_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_420_to_fp16, x = inputs_7_cast_fp16)[name = tensor("out_7_cast_fp16")]; tensor obj_15_gamma_0_to_fp16 = const()[name = tensor("obj_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186400576)))]; tensor obj_15_beta_0_to_fp16 = const()[name = tensor("obj_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186403200)))]; tensor obj_15_epsilon_0_to_fp16 = const()[name = tensor("obj_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_15_cast_fp16 = batch_norm(beta = obj_15_beta_0_to_fp16, epsilon = obj_15_epsilon_0_to_fp16, gamma = obj_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = tensor("obj_15_cast_fp16")]; tensor query_5_pad_type_0 = const()[name = tensor("query_5_pad_type_0"), val = tensor("valid")]; tensor query_5_strides_0 = const()[name = tensor("query_5_strides_0"), val = tensor([1, 1])]; tensor query_5_pad_0 = const()[name = tensor("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_5_dilations_0 = const()[name = tensor("query_5_dilations_0"), val = tensor([1, 1])]; tensor query_5_groups_0 = const()[name = tensor("query_5_groups_0"), val = tensor(1)]; tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186405824)))]; tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189682688)))]; tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_15_cast_fp16)[name = tensor("query_5_cast_fp16")]; tensor current_key_3_pad_type_0 = const()[name = tensor("current_key_3_pad_type_0"), val = tensor("valid")]; tensor current_key_3_strides_0 = const()[name = tensor("current_key_3_strides_0"), val = tensor([1, 1])]; tensor current_key_3_pad_0 = const()[name = tensor("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_3_dilations_0 = const()[name = tensor("current_key_3_dilations_0"), val = tensor([1, 1])]; tensor current_key_3_groups_0 = const()[name = tensor("current_key_3_groups_0"), val = tensor(1)]; tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189685312)))]; tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_15_cast_fp16)[name = tensor("current_key_3_cast_fp16")]; tensor current_value_3_pad_type_0 = const()[name = tensor("current_value_3_pad_type_0"), val = tensor("valid")]; tensor current_value_3_strides_0 = const()[name = tensor("current_value_3_strides_0"), val = tensor([1, 1])]; tensor current_value_3_pad_0 = const()[name = tensor("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_3_dilations_0 = const()[name = tensor("current_value_3_dilations_0"), val = tensor([1, 1])]; tensor current_value_3_groups_0 = const()[name = tensor("current_value_3_groups_0"), val = tensor(1)]; tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192962176)))]; tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196239040)))]; tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_15_cast_fp16)[name = tensor("current_value_3_cast_fp16")]; tensor var_459_cast_fp16 = mul(x = var_103_cast_fp16_1, y = var_239_cast_fp16)[name = tensor("op_459_cast_fp16")]; tensor var_460_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_460_cast_fp16")]; tensor key_5_cast_fp16 = add(x = var_459_cast_fp16, y = var_460_cast_fp16)[name = tensor("key_5_cast_fp16")]; tensor var_463_cast_fp16 = mul(x = var_138_cast_fp16_1, y = var_239_cast_fp16)[name = tensor("op_463_cast_fp16")]; tensor var_464_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_464_cast_fp16")]; tensor value_5_cast_fp16 = add(x = var_463_cast_fp16, y = var_464_cast_fp16)[name = tensor("value_5_cast_fp16")]; tensor var_468 = const()[name = tensor("op_468"), val = tensor([1, 20, 64, 1])]; tensor mh_q_5_cast_fp16 = reshape(shape = var_468, x = query_5_cast_fp16)[name = tensor("mh_q_5_cast_fp16")]; tensor var_470_to_fp16 = const()[name = tensor("op_470_to_fp16"), val = tensor(0x1p-3)]; tensor var_471_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_470_to_fp16)[name = tensor("op_471_cast_fp16")]; tensor var_474 = const()[name = tensor("op_474"), val = tensor([1, 20, 64, 448])]; tensor var_475_cast_fp16 = reshape(shape = var_474, x = key_5_cast_fp16)[name = tensor("op_475_cast_fp16")]; tensor mh_w_7_transpose_x_0 = const()[name = tensor("mh_w_7_transpose_x_0"), val = tensor(true)]; tensor mh_w_7_transpose_y_0 = const()[name = tensor("mh_w_7_transpose_y_0"), val = tensor(false)]; tensor mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_471_cast_fp16, y = var_475_cast_fp16)[name = tensor("mh_w_7_cast_fp16")]; tensor mh_w_9_cast_fp16 = add(x = mh_w_7_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_9_cast_fp16")]; tensor var_483_cast_fp16 = softmax(axis = var_395, x = mh_w_9_cast_fp16)[name = tensor("op_483_cast_fp16")]; tensor var_484 = const()[name = tensor("op_484"), val = tensor([1, 20, 64, 448])]; tensor var_485_cast_fp16 = reshape(shape = var_484, x = value_5_cast_fp16)[name = tensor("op_485_cast_fp16")]; tensor attn_5_transpose_x_0 = const()[name = tensor("attn_5_transpose_x_0"), val = tensor(false)]; tensor attn_5_transpose_y_0 = const()[name = tensor("attn_5_transpose_y_0"), val = tensor(true)]; tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_485_cast_fp16, y = var_483_cast_fp16)[name = tensor("attn_5_cast_fp16")]; tensor var_488 = const()[name = tensor("op_488"), val = tensor([1, 1280, 1, 1])]; tensor input_11_cast_fp16 = reshape(shape = var_488, x = attn_5_cast_fp16)[name = tensor("input_11_cast_fp16")]; tensor obj_21_pad_type_0 = const()[name = tensor("obj_21_pad_type_0"), val = tensor("valid")]; tensor obj_21_strides_0 = const()[name = tensor("obj_21_strides_0"), val = tensor([1, 1])]; tensor obj_21_pad_0 = const()[name = tensor("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_21_dilations_0 = const()[name = tensor("obj_21_dilations_0"), val = tensor([1, 1])]; tensor obj_21_groups_0 = const()[name = tensor("obj_21_groups_0"), val = tensor(1)]; tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196241664)))]; tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199518528)))]; tensor obj_21_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = tensor("obj_21_cast_fp16")]; tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_21_cast_fp16)[name = tensor("inputs_9_cast_fp16")]; tensor out_9_axes_0 = const()[name = tensor("out_9_axes_0"), val = tensor([1])]; tensor var_510_to_fp16 = const()[name = tensor("op_510_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_510_to_fp16, x = inputs_9_cast_fp16)[name = tensor("out_9_cast_fp16")]; tensor obj_23_gamma_0_to_fp16 = const()[name = tensor("obj_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199521152)))]; tensor obj_23_beta_0_to_fp16 = const()[name = tensor("obj_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199523776)))]; tensor obj_23_epsilon_0_to_fp16 = const()[name = tensor("obj_23_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_23_cast_fp16 = batch_norm(beta = obj_23_beta_0_to_fp16, epsilon = obj_23_epsilon_0_to_fp16, gamma = obj_23_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = tensor("obj_23_cast_fp16")]; tensor query_7_pad_type_0 = const()[name = tensor("query_7_pad_type_0"), val = tensor("valid")]; tensor query_7_strides_0 = const()[name = tensor("query_7_strides_0"), val = tensor([1, 1])]; tensor query_7_pad_0 = const()[name = tensor("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_7_dilations_0 = const()[name = tensor("query_7_dilations_0"), val = tensor([1, 1])]; tensor query_7_groups_0 = const()[name = tensor("query_7_groups_0"), val = tensor(1)]; tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199526400)))]; tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202803264)))]; tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_23_cast_fp16)[name = tensor("query_7_cast_fp16")]; tensor key_7_pad_type_0 = const()[name = tensor("key_7_pad_type_0"), val = tensor("valid")]; tensor key_7_strides_0 = const()[name = tensor("key_7_strides_0"), val = tensor([1, 1])]; tensor key_7_pad_0 = const()[name = tensor("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_7_dilations_0 = const()[name = tensor("key_7_dilations_0"), val = tensor([1, 1])]; tensor key_7_groups_0 = const()[name = tensor("key_7_groups_0"), val = tensor(1)]; tensor layers_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202805888)))]; tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_7_cast_fp16")]; tensor value_7_pad_type_0 = const()[name = tensor("value_7_pad_type_0"), val = tensor("valid")]; tensor value_7_strides_0 = const()[name = tensor("value_7_strides_0"), val = tensor([1, 1])]; tensor value_7_pad_0 = const()[name = tensor("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_7_dilations_0 = const()[name = tensor("value_7_dilations_0"), val = tensor([1, 1])]; tensor value_7_groups_0 = const()[name = tensor("value_7_groups_0"), val = tensor(1)]; tensor layers_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206082752)))]; tensor layers_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209359616)))]; tensor value_7_cast_fp16 = conv(bias = layers_1_encoder_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_7_cast_fp16")]; tensor var_546 = const()[name = tensor("op_546"), val = tensor([1, 20, 64, 1])]; tensor mh_q_7_cast_fp16 = reshape(shape = var_546, x = query_7_cast_fp16)[name = tensor("mh_q_7_cast_fp16")]; tensor var_548_to_fp16 = const()[name = tensor("op_548_to_fp16"), val = tensor(0x1p-3)]; tensor var_549_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_548_to_fp16)[name = tensor("op_549_cast_fp16")]; tensor var_552 = const()[name = tensor("op_552"), val = tensor([1, 20, 64, 1500])]; tensor var_553_cast_fp16 = reshape(shape = var_552, x = key_7_cast_fp16)[name = tensor("op_553_cast_fp16")]; tensor mh_w_11_transpose_x_0 = const()[name = tensor("mh_w_11_transpose_x_0"), val = tensor(true)]; tensor mh_w_11_transpose_y_0 = const()[name = tensor("mh_w_11_transpose_y_0"), val = tensor(false)]; tensor mh_w_11_cast_fp16 = matmul(transpose_x = mh_w_11_transpose_x_0, transpose_y = mh_w_11_transpose_y_0, x = var_549_cast_fp16, y = var_553_cast_fp16)[name = tensor("mh_w_11_cast_fp16")]; tensor obj_27_cast_fp16 = softmax(axis = var_395, x = mh_w_11_cast_fp16)[name = tensor("obj_27_cast_fp16")]; tensor var_557 = const()[name = tensor("op_557"), val = tensor([1, 20, 64, 1500])]; tensor var_558_cast_fp16 = reshape(shape = var_557, x = value_7_cast_fp16)[name = tensor("op_558_cast_fp16")]; tensor attn_7_transpose_x_0 = const()[name = tensor("attn_7_transpose_x_0"), val = tensor(false)]; tensor attn_7_transpose_y_0 = const()[name = tensor("attn_7_transpose_y_0"), val = tensor(true)]; tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_558_cast_fp16, y = obj_27_cast_fp16)[name = tensor("attn_7_cast_fp16")]; tensor var_561 = const()[name = tensor("op_561"), val = tensor([1, 1280, 1, 1])]; tensor input_13_cast_fp16 = reshape(shape = var_561, x = attn_7_cast_fp16)[name = tensor("input_13_cast_fp16")]; tensor obj_25_pad_type_0 = const()[name = tensor("obj_25_pad_type_0"), val = tensor("valid")]; tensor obj_25_strides_0 = const()[name = tensor("obj_25_strides_0"), val = tensor([1, 1])]; tensor obj_25_pad_0 = const()[name = tensor("obj_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_25_dilations_0 = const()[name = tensor("obj_25_dilations_0"), val = tensor([1, 1])]; tensor obj_25_groups_0 = const()[name = tensor("obj_25_groups_0"), val = tensor(1)]; tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209362240)))]; tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212639104)))]; tensor obj_25_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_25_dilations_0, groups = obj_25_groups_0, pad = obj_25_pad_0, pad_type = obj_25_pad_type_0, strides = obj_25_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = tensor("obj_25_cast_fp16")]; tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_25_cast_fp16)[name = tensor("inputs_11_cast_fp16")]; tensor out_11_axes_0 = const()[name = tensor("out_11_axes_0"), val = tensor([1])]; tensor var_579_to_fp16 = const()[name = tensor("op_579_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_579_to_fp16, x = inputs_11_cast_fp16)[name = tensor("out_11_cast_fp16")]; tensor input_15_gamma_0_to_fp16 = const()[name = tensor("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212641728)))]; tensor input_15_beta_0_to_fp16 = const()[name = tensor("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212644352)))]; tensor input_15_epsilon_0_to_fp16 = const()[name = tensor("input_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = tensor("input_15_cast_fp16")]; tensor input_17_pad_type_0 = const()[name = tensor("input_17_pad_type_0"), val = tensor("valid")]; tensor input_17_strides_0 = const()[name = tensor("input_17_strides_0"), val = tensor([1, 1])]; tensor input_17_pad_0 = const()[name = tensor("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_17_dilations_0 = const()[name = tensor("input_17_dilations_0"), val = tensor([1, 1])]; tensor input_17_groups_0 = const()[name = tensor("input_17_groups_0"), val = tensor(1)]; tensor layers_1_fc1_weight_to_fp16 = const()[name = tensor("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212646976)))]; tensor layers_1_fc1_bias_to_fp16 = const()[name = tensor("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225754240)))]; tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = tensor("input_17_cast_fp16")]; tensor input_19_mode_0 = const()[name = tensor("input_19_mode_0"), val = tensor("EXACT")]; tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = tensor("input_19_cast_fp16")]; tensor hidden_states_5_pad_type_0 = const()[name = tensor("hidden_states_5_pad_type_0"), val = tensor("valid")]; tensor hidden_states_5_strides_0 = const()[name = tensor("hidden_states_5_strides_0"), val = tensor([1, 1])]; tensor hidden_states_5_pad_0 = const()[name = tensor("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_5_dilations_0 = const()[name = tensor("hidden_states_5_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_5_groups_0 = const()[name = tensor("hidden_states_5_groups_0"), val = tensor(1)]; tensor layers_1_fc2_weight_to_fp16 = const()[name = tensor("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225764544)))]; tensor layers_1_fc2_bias_to_fp16 = const()[name = tensor("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238871808)))]; tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = tensor("hidden_states_5_cast_fp16")]; tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor("inputs_13_cast_fp16")]; tensor var_614 = const()[name = tensor("op_614"), val = tensor(3)]; tensor out_13_axes_0 = const()[name = tensor("out_13_axes_0"), val = tensor([1])]; tensor var_639_to_fp16 = const()[name = tensor("op_639_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_639_to_fp16, x = inputs_13_cast_fp16)[name = tensor("out_13_cast_fp16")]; tensor obj_29_gamma_0_to_fp16 = const()[name = tensor("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238874432)))]; tensor obj_29_beta_0_to_fp16 = const()[name = tensor("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238877056)))]; tensor obj_29_epsilon_0_to_fp16 = const()[name = tensor("obj_29_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = tensor("obj_29_cast_fp16")]; tensor query_9_pad_type_0 = const()[name = tensor("query_9_pad_type_0"), val = tensor("valid")]; tensor query_9_strides_0 = const()[name = tensor("query_9_strides_0"), val = tensor([1, 1])]; tensor query_9_pad_0 = const()[name = tensor("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_9_dilations_0 = const()[name = tensor("query_9_dilations_0"), val = tensor([1, 1])]; tensor query_9_groups_0 = const()[name = tensor("query_9_groups_0"), val = tensor(1)]; tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238879680)))]; tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242156544)))]; tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor("query_9_cast_fp16")]; tensor current_key_5_pad_type_0 = const()[name = tensor("current_key_5_pad_type_0"), val = tensor("valid")]; tensor current_key_5_strides_0 = const()[name = tensor("current_key_5_strides_0"), val = tensor([1, 1])]; tensor current_key_5_pad_0 = const()[name = tensor("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_5_dilations_0 = const()[name = tensor("current_key_5_dilations_0"), val = tensor([1, 1])]; tensor current_key_5_groups_0 = const()[name = tensor("current_key_5_groups_0"), val = tensor(1)]; tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242159168)))]; tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor("current_key_5_cast_fp16")]; tensor current_value_5_pad_type_0 = const()[name = tensor("current_value_5_pad_type_0"), val = tensor("valid")]; tensor current_value_5_strides_0 = const()[name = tensor("current_value_5_strides_0"), val = tensor([1, 1])]; tensor current_value_5_pad_0 = const()[name = tensor("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_5_dilations_0 = const()[name = tensor("current_value_5_dilations_0"), val = tensor([1, 1])]; tensor current_value_5_groups_0 = const()[name = tensor("current_value_5_groups_0"), val = tensor(1)]; tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245436032)))]; tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248712896)))]; tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor("current_value_5_cast_fp16")]; tensor var_678_cast_fp16 = mul(x = var_103_cast_fp16_2, y = var_239_cast_fp16)[name = tensor("op_678_cast_fp16")]; tensor var_679_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_679_cast_fp16")]; tensor key_9_cast_fp16 = add(x = var_678_cast_fp16, y = var_679_cast_fp16)[name = tensor("key_9_cast_fp16")]; tensor var_682_cast_fp16 = mul(x = var_138_cast_fp16_2, y = var_239_cast_fp16)[name = tensor("op_682_cast_fp16")]; tensor var_683_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_683_cast_fp16")]; tensor value_9_cast_fp16 = add(x = var_682_cast_fp16, y = var_683_cast_fp16)[name = tensor("value_9_cast_fp16")]; tensor var_687 = const()[name = tensor("op_687"), val = tensor([1, 20, 64, 1])]; tensor mh_q_9_cast_fp16 = reshape(shape = var_687, x = query_9_cast_fp16)[name = tensor("mh_q_9_cast_fp16")]; tensor var_689_to_fp16 = const()[name = tensor("op_689_to_fp16"), val = tensor(0x1p-3)]; tensor var_690_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_689_to_fp16)[name = tensor("op_690_cast_fp16")]; tensor var_693 = const()[name = tensor("op_693"), val = tensor([1, 20, 64, 448])]; tensor var_694_cast_fp16 = reshape(shape = var_693, x = key_9_cast_fp16)[name = tensor("op_694_cast_fp16")]; tensor mh_w_13_transpose_x_0 = const()[name = tensor("mh_w_13_transpose_x_0"), val = tensor(true)]; tensor mh_w_13_transpose_y_0 = const()[name = tensor("mh_w_13_transpose_y_0"), val = tensor(false)]; tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_690_cast_fp16, y = var_694_cast_fp16)[name = tensor("mh_w_13_cast_fp16")]; tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_15_cast_fp16")]; tensor var_702_cast_fp16 = softmax(axis = var_614, x = mh_w_15_cast_fp16)[name = tensor("op_702_cast_fp16")]; tensor var_703 = const()[name = tensor("op_703"), val = tensor([1, 20, 64, 448])]; tensor var_704_cast_fp16 = reshape(shape = var_703, x = value_9_cast_fp16)[name = tensor("op_704_cast_fp16")]; tensor attn_9_transpose_x_0 = const()[name = tensor("attn_9_transpose_x_0"), val = tensor(false)]; tensor attn_9_transpose_y_0 = const()[name = tensor("attn_9_transpose_y_0"), val = tensor(true)]; tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_704_cast_fp16, y = var_702_cast_fp16)[name = tensor("attn_9_cast_fp16")]; tensor var_707 = const()[name = tensor("op_707"), val = tensor([1, 1280, 1, 1])]; tensor input_21_cast_fp16 = reshape(shape = var_707, x = attn_9_cast_fp16)[name = tensor("input_21_cast_fp16")]; tensor obj_35_pad_type_0 = const()[name = tensor("obj_35_pad_type_0"), val = tensor("valid")]; tensor obj_35_strides_0 = const()[name = tensor("obj_35_strides_0"), val = tensor([1, 1])]; tensor obj_35_pad_0 = const()[name = tensor("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_35_dilations_0 = const()[name = tensor("obj_35_dilations_0"), val = tensor([1, 1])]; tensor obj_35_groups_0 = const()[name = tensor("obj_35_groups_0"), val = tensor(1)]; tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248715520)))]; tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251992384)))]; tensor obj_35_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor("obj_35_cast_fp16")]; tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_35_cast_fp16)[name = tensor("inputs_15_cast_fp16")]; tensor out_15_axes_0 = const()[name = tensor("out_15_axes_0"), val = tensor([1])]; tensor var_729_to_fp16 = const()[name = tensor("op_729_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_729_to_fp16, x = inputs_15_cast_fp16)[name = tensor("out_15_cast_fp16")]; tensor obj_37_gamma_0_to_fp16 = const()[name = tensor("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251995008)))]; tensor obj_37_beta_0_to_fp16 = const()[name = tensor("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251997632)))]; tensor obj_37_epsilon_0_to_fp16 = const()[name = tensor("obj_37_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = tensor("obj_37_cast_fp16")]; tensor query_11_pad_type_0 = const()[name = tensor("query_11_pad_type_0"), val = tensor("valid")]; tensor query_11_strides_0 = const()[name = tensor("query_11_strides_0"), val = tensor([1, 1])]; tensor query_11_pad_0 = const()[name = tensor("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_11_dilations_0 = const()[name = tensor("query_11_dilations_0"), val = tensor([1, 1])]; tensor query_11_groups_0 = const()[name = tensor("query_11_groups_0"), val = tensor(1)]; tensor layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252000256)))]; tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(255277120)))]; tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor("query_11_cast_fp16")]; tensor key_11_pad_type_0 = const()[name = tensor("key_11_pad_type_0"), val = tensor("valid")]; tensor key_11_strides_0 = const()[name = tensor("key_11_strides_0"), val = tensor([1, 1])]; tensor key_11_pad_0 = const()[name = tensor("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_11_dilations_0 = const()[name = tensor("key_11_dilations_0"), val = tensor([1, 1])]; tensor key_11_groups_0 = const()[name = tensor("key_11_groups_0"), val = tensor(1)]; tensor layers_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(255279744)))]; tensor key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_11_cast_fp16")]; tensor value_11_pad_type_0 = const()[name = tensor("value_11_pad_type_0"), val = tensor("valid")]; tensor value_11_strides_0 = const()[name = tensor("value_11_strides_0"), val = tensor([1, 1])]; tensor value_11_pad_0 = const()[name = tensor("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_11_dilations_0 = const()[name = tensor("value_11_dilations_0"), val = tensor([1, 1])]; tensor value_11_groups_0 = const()[name = tensor("value_11_groups_0"), val = tensor(1)]; tensor layers_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258556608)))]; tensor layers_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261833472)))]; tensor value_11_cast_fp16 = conv(bias = layers_2_encoder_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_11_cast_fp16")]; tensor var_765 = const()[name = tensor("op_765"), val = tensor([1, 20, 64, 1])]; tensor mh_q_11_cast_fp16 = reshape(shape = var_765, x = query_11_cast_fp16)[name = tensor("mh_q_11_cast_fp16")]; tensor var_767_to_fp16 = const()[name = tensor("op_767_to_fp16"), val = tensor(0x1p-3)]; tensor var_768_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_767_to_fp16)[name = tensor("op_768_cast_fp16")]; tensor var_771 = const()[name = tensor("op_771"), val = tensor([1, 20, 64, 1500])]; tensor var_772_cast_fp16 = reshape(shape = var_771, x = key_11_cast_fp16)[name = tensor("op_772_cast_fp16")]; tensor mh_w_17_transpose_x_0 = const()[name = tensor("mh_w_17_transpose_x_0"), val = tensor(true)]; tensor mh_w_17_transpose_y_0 = const()[name = tensor("mh_w_17_transpose_y_0"), val = tensor(false)]; tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_768_cast_fp16, y = var_772_cast_fp16)[name = tensor("mh_w_17_cast_fp16")]; tensor obj_41_cast_fp16 = softmax(axis = var_614, x = mh_w_17_cast_fp16)[name = tensor("obj_41_cast_fp16")]; tensor var_776 = const()[name = tensor("op_776"), val = tensor([1, 20, 64, 1500])]; tensor var_777_cast_fp16 = reshape(shape = var_776, x = value_11_cast_fp16)[name = tensor("op_777_cast_fp16")]; tensor attn_11_transpose_x_0 = const()[name = tensor("attn_11_transpose_x_0"), val = tensor(false)]; tensor attn_11_transpose_y_0 = const()[name = tensor("attn_11_transpose_y_0"), val = tensor(true)]; tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_777_cast_fp16, y = obj_41_cast_fp16)[name = tensor("attn_11_cast_fp16")]; tensor var_780 = const()[name = tensor("op_780"), val = tensor([1, 1280, 1, 1])]; tensor input_23_cast_fp16 = reshape(shape = var_780, x = attn_11_cast_fp16)[name = tensor("input_23_cast_fp16")]; tensor obj_39_pad_type_0 = const()[name = tensor("obj_39_pad_type_0"), val = tensor("valid")]; tensor obj_39_strides_0 = const()[name = tensor("obj_39_strides_0"), val = tensor([1, 1])]; tensor obj_39_pad_0 = const()[name = tensor("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_39_dilations_0 = const()[name = tensor("obj_39_dilations_0"), val = tensor([1, 1])]; tensor obj_39_groups_0 = const()[name = tensor("obj_39_groups_0"), val = tensor(1)]; tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261836096)))]; tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265112960)))]; tensor obj_39_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = tensor("obj_39_cast_fp16")]; tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_39_cast_fp16)[name = tensor("inputs_17_cast_fp16")]; tensor out_17_axes_0 = const()[name = tensor("out_17_axes_0"), val = tensor([1])]; tensor var_798_to_fp16 = const()[name = tensor("op_798_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_798_to_fp16, x = inputs_17_cast_fp16)[name = tensor("out_17_cast_fp16")]; tensor input_25_gamma_0_to_fp16 = const()[name = tensor("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265115584)))]; tensor input_25_beta_0_to_fp16 = const()[name = tensor("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265118208)))]; tensor input_25_epsilon_0_to_fp16 = const()[name = tensor("input_25_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = tensor("input_25_cast_fp16")]; tensor input_27_pad_type_0 = const()[name = tensor("input_27_pad_type_0"), val = tensor("valid")]; tensor input_27_strides_0 = const()[name = tensor("input_27_strides_0"), val = tensor([1, 1])]; tensor input_27_pad_0 = const()[name = tensor("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_27_dilations_0 = const()[name = tensor("input_27_dilations_0"), val = tensor([1, 1])]; tensor input_27_groups_0 = const()[name = tensor("input_27_groups_0"), val = tensor(1)]; tensor layers_2_fc1_weight_to_fp16 = const()[name = tensor("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265120832)))]; tensor layers_2_fc1_bias_to_fp16 = const()[name = tensor("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278228096)))]; tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = tensor("input_27_cast_fp16")]; tensor input_29_mode_0 = const()[name = tensor("input_29_mode_0"), val = tensor("EXACT")]; tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = tensor("input_29_cast_fp16")]; tensor hidden_states_7_pad_type_0 = const()[name = tensor("hidden_states_7_pad_type_0"), val = tensor("valid")]; tensor hidden_states_7_strides_0 = const()[name = tensor("hidden_states_7_strides_0"), val = tensor([1, 1])]; tensor hidden_states_7_pad_0 = const()[name = tensor("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_7_dilations_0 = const()[name = tensor("hidden_states_7_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_7_groups_0 = const()[name = tensor("hidden_states_7_groups_0"), val = tensor(1)]; tensor layers_2_fc2_weight_to_fp16 = const()[name = tensor("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278238400)))]; tensor layers_2_fc2_bias_to_fp16 = const()[name = tensor("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291345664)))]; tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = tensor("inputs_19_cast_fp16")]; tensor var_833 = const()[name = tensor("op_833"), val = tensor(3)]; tensor out_19_axes_0 = const()[name = tensor("out_19_axes_0"), val = tensor([1])]; tensor var_858_to_fp16 = const()[name = tensor("op_858_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_858_to_fp16, x = inputs_19_cast_fp16)[name = tensor("out_19_cast_fp16")]; tensor obj_43_gamma_0_to_fp16 = const()[name = tensor("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291348288)))]; tensor obj_43_beta_0_to_fp16 = const()[name = tensor("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291350912)))]; tensor obj_43_epsilon_0_to_fp16 = const()[name = tensor("obj_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = tensor("obj_43_cast_fp16")]; tensor query_13_pad_type_0 = const()[name = tensor("query_13_pad_type_0"), val = tensor("valid")]; tensor query_13_strides_0 = const()[name = tensor("query_13_strides_0"), val = tensor([1, 1])]; tensor query_13_pad_0 = const()[name = tensor("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_13_dilations_0 = const()[name = tensor("query_13_dilations_0"), val = tensor([1, 1])]; tensor query_13_groups_0 = const()[name = tensor("query_13_groups_0"), val = tensor(1)]; tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291353536)))]; tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294630400)))]; tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = tensor("query_13_cast_fp16")]; tensor current_key_7_pad_type_0 = const()[name = tensor("current_key_7_pad_type_0"), val = tensor("valid")]; tensor current_key_7_strides_0 = const()[name = tensor("current_key_7_strides_0"), val = tensor([1, 1])]; tensor current_key_7_pad_0 = const()[name = tensor("current_key_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_7_dilations_0 = const()[name = tensor("current_key_7_dilations_0"), val = tensor([1, 1])]; tensor current_key_7_groups_0 = const()[name = tensor("current_key_7_groups_0"), val = tensor(1)]; tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(294633024)))]; tensor current_key_7_cast_fp16 = conv(dilations = current_key_7_dilations_0, groups = current_key_7_groups_0, pad = current_key_7_pad_0, pad_type = current_key_7_pad_type_0, strides = current_key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = tensor("current_key_7_cast_fp16")]; tensor current_value_7_pad_type_0 = const()[name = tensor("current_value_7_pad_type_0"), val = tensor("valid")]; tensor current_value_7_strides_0 = const()[name = tensor("current_value_7_strides_0"), val = tensor([1, 1])]; tensor current_value_7_pad_0 = const()[name = tensor("current_value_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_7_dilations_0 = const()[name = tensor("current_value_7_dilations_0"), val = tensor([1, 1])]; tensor current_value_7_groups_0 = const()[name = tensor("current_value_7_groups_0"), val = tensor(1)]; tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297909888)))]; tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301186752)))]; tensor current_value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_7_dilations_0, groups = current_value_7_groups_0, pad = current_value_7_pad_0, pad_type = current_value_7_pad_type_0, strides = current_value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = tensor("current_value_7_cast_fp16")]; tensor var_897_cast_fp16 = mul(x = var_103_cast_fp16_3, y = var_239_cast_fp16)[name = tensor("op_897_cast_fp16")]; tensor var_898_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_898_cast_fp16")]; tensor key_13_cast_fp16 = add(x = var_897_cast_fp16, y = var_898_cast_fp16)[name = tensor("key_13_cast_fp16")]; tensor var_901_cast_fp16 = mul(x = var_138_cast_fp16_3, y = var_239_cast_fp16)[name = tensor("op_901_cast_fp16")]; tensor var_902_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_902_cast_fp16")]; tensor value_13_cast_fp16 = add(x = var_901_cast_fp16, y = var_902_cast_fp16)[name = tensor("value_13_cast_fp16")]; tensor var_906 = const()[name = tensor("op_906"), val = tensor([1, 20, 64, 1])]; tensor mh_q_13_cast_fp16 = reshape(shape = var_906, x = query_13_cast_fp16)[name = tensor("mh_q_13_cast_fp16")]; tensor var_908_to_fp16 = const()[name = tensor("op_908_to_fp16"), val = tensor(0x1p-3)]; tensor var_909_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_908_to_fp16)[name = tensor("op_909_cast_fp16")]; tensor var_912 = const()[name = tensor("op_912"), val = tensor([1, 20, 64, 448])]; tensor var_913_cast_fp16 = reshape(shape = var_912, x = key_13_cast_fp16)[name = tensor("op_913_cast_fp16")]; tensor mh_w_19_transpose_x_0 = const()[name = tensor("mh_w_19_transpose_x_0"), val = tensor(true)]; tensor mh_w_19_transpose_y_0 = const()[name = tensor("mh_w_19_transpose_y_0"), val = tensor(false)]; tensor mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_909_cast_fp16, y = var_913_cast_fp16)[name = tensor("mh_w_19_cast_fp16")]; tensor mh_w_21_cast_fp16 = add(x = mh_w_19_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_21_cast_fp16")]; tensor var_921_cast_fp16 = softmax(axis = var_833, x = mh_w_21_cast_fp16)[name = tensor("op_921_cast_fp16")]; tensor var_922 = const()[name = tensor("op_922"), val = tensor([1, 20, 64, 448])]; tensor var_923_cast_fp16 = reshape(shape = var_922, x = value_13_cast_fp16)[name = tensor("op_923_cast_fp16")]; tensor attn_13_transpose_x_0 = const()[name = tensor("attn_13_transpose_x_0"), val = tensor(false)]; tensor attn_13_transpose_y_0 = const()[name = tensor("attn_13_transpose_y_0"), val = tensor(true)]; tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_923_cast_fp16, y = var_921_cast_fp16)[name = tensor("attn_13_cast_fp16")]; tensor var_926 = const()[name = tensor("op_926"), val = tensor([1, 1280, 1, 1])]; tensor input_31_cast_fp16 = reshape(shape = var_926, x = attn_13_cast_fp16)[name = tensor("input_31_cast_fp16")]; tensor obj_49_pad_type_0 = const()[name = tensor("obj_49_pad_type_0"), val = tensor("valid")]; tensor obj_49_strides_0 = const()[name = tensor("obj_49_strides_0"), val = tensor([1, 1])]; tensor obj_49_pad_0 = const()[name = tensor("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_49_dilations_0 = const()[name = tensor("obj_49_dilations_0"), val = tensor([1, 1])]; tensor obj_49_groups_0 = const()[name = tensor("obj_49_groups_0"), val = tensor(1)]; tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301189376)))]; tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304466240)))]; tensor obj_49_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = tensor("obj_49_cast_fp16")]; tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_49_cast_fp16)[name = tensor("inputs_21_cast_fp16")]; tensor out_21_axes_0 = const()[name = tensor("out_21_axes_0"), val = tensor([1])]; tensor var_948_to_fp16 = const()[name = tensor("op_948_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_948_to_fp16, x = inputs_21_cast_fp16)[name = tensor("out_21_cast_fp16")]; tensor obj_51_gamma_0_to_fp16 = const()[name = tensor("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304468864)))]; tensor obj_51_beta_0_to_fp16 = const()[name = tensor("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304471488)))]; tensor obj_51_epsilon_0_to_fp16 = const()[name = tensor("obj_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = tensor("obj_51_cast_fp16")]; tensor query_15_pad_type_0 = const()[name = tensor("query_15_pad_type_0"), val = tensor("valid")]; tensor query_15_strides_0 = const()[name = tensor("query_15_strides_0"), val = tensor([1, 1])]; tensor query_15_pad_0 = const()[name = tensor("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_15_dilations_0 = const()[name = tensor("query_15_dilations_0"), val = tensor([1, 1])]; tensor query_15_groups_0 = const()[name = tensor("query_15_groups_0"), val = tensor(1)]; tensor layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304474112)))]; tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307750976)))]; tensor query_15_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = tensor("query_15_cast_fp16")]; tensor key_15_pad_type_0 = const()[name = tensor("key_15_pad_type_0"), val = tensor("valid")]; tensor key_15_strides_0 = const()[name = tensor("key_15_strides_0"), val = tensor([1, 1])]; tensor key_15_pad_0 = const()[name = tensor("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_15_dilations_0 = const()[name = tensor("key_15_dilations_0"), val = tensor([1, 1])]; tensor key_15_groups_0 = const()[name = tensor("key_15_groups_0"), val = tensor(1)]; tensor layers_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307753600)))]; tensor key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_15_cast_fp16")]; tensor value_15_pad_type_0 = const()[name = tensor("value_15_pad_type_0"), val = tensor("valid")]; tensor value_15_strides_0 = const()[name = tensor("value_15_strides_0"), val = tensor([1, 1])]; tensor value_15_pad_0 = const()[name = tensor("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_15_dilations_0 = const()[name = tensor("value_15_dilations_0"), val = tensor([1, 1])]; tensor value_15_groups_0 = const()[name = tensor("value_15_groups_0"), val = tensor(1)]; tensor layers_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311030464)))]; tensor layers_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314307328)))]; tensor value_15_cast_fp16 = conv(bias = layers_3_encoder_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_15_cast_fp16")]; tensor var_984 = const()[name = tensor("op_984"), val = tensor([1, 20, 64, 1])]; tensor mh_q_15_cast_fp16 = reshape(shape = var_984, x = query_15_cast_fp16)[name = tensor("mh_q_15_cast_fp16")]; tensor var_986_to_fp16 = const()[name = tensor("op_986_to_fp16"), val = tensor(0x1p-3)]; tensor var_987_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_986_to_fp16)[name = tensor("op_987_cast_fp16")]; tensor var_990 = const()[name = tensor("op_990"), val = tensor([1, 20, 64, 1500])]; tensor var_991_cast_fp16 = reshape(shape = var_990, x = key_15_cast_fp16)[name = tensor("op_991_cast_fp16")]; tensor mh_w_23_transpose_x_0 = const()[name = tensor("mh_w_23_transpose_x_0"), val = tensor(true)]; tensor mh_w_23_transpose_y_0 = const()[name = tensor("mh_w_23_transpose_y_0"), val = tensor(false)]; tensor mh_w_23_cast_fp16 = matmul(transpose_x = mh_w_23_transpose_x_0, transpose_y = mh_w_23_transpose_y_0, x = var_987_cast_fp16, y = var_991_cast_fp16)[name = tensor("mh_w_23_cast_fp16")]; tensor obj_55_cast_fp16 = softmax(axis = var_833, x = mh_w_23_cast_fp16)[name = tensor("obj_55_cast_fp16")]; tensor var_995 = const()[name = tensor("op_995"), val = tensor([1, 20, 64, 1500])]; tensor var_996_cast_fp16 = reshape(shape = var_995, x = value_15_cast_fp16)[name = tensor("op_996_cast_fp16")]; tensor attn_15_transpose_x_0 = const()[name = tensor("attn_15_transpose_x_0"), val = tensor(false)]; tensor attn_15_transpose_y_0 = const()[name = tensor("attn_15_transpose_y_0"), val = tensor(true)]; tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_996_cast_fp16, y = obj_55_cast_fp16)[name = tensor("attn_15_cast_fp16")]; tensor var_999 = const()[name = tensor("op_999"), val = tensor([1, 1280, 1, 1])]; tensor input_33_cast_fp16 = reshape(shape = var_999, x = attn_15_cast_fp16)[name = tensor("input_33_cast_fp16")]; tensor obj_53_pad_type_0 = const()[name = tensor("obj_53_pad_type_0"), val = tensor("valid")]; tensor obj_53_strides_0 = const()[name = tensor("obj_53_strides_0"), val = tensor([1, 1])]; tensor obj_53_pad_0 = const()[name = tensor("obj_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_53_dilations_0 = const()[name = tensor("obj_53_dilations_0"), val = tensor([1, 1])]; tensor obj_53_groups_0 = const()[name = tensor("obj_53_groups_0"), val = tensor(1)]; tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314309952)))]; tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(317586816)))]; tensor obj_53_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_53_dilations_0, groups = obj_53_groups_0, pad = obj_53_pad_0, pad_type = obj_53_pad_type_0, strides = obj_53_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = tensor("obj_53_cast_fp16")]; tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_53_cast_fp16)[name = tensor("inputs_23_cast_fp16")]; tensor out_23_axes_0 = const()[name = tensor("out_23_axes_0"), val = tensor([1])]; tensor var_1017_to_fp16 = const()[name = tensor("op_1017_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_1017_to_fp16, x = inputs_23_cast_fp16)[name = tensor("out_23_cast_fp16")]; tensor input_35_gamma_0_to_fp16 = const()[name = tensor("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(317589440)))]; tensor input_35_beta_0_to_fp16 = const()[name = tensor("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(317592064)))]; tensor input_35_epsilon_0_to_fp16 = const()[name = tensor("input_35_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = tensor("input_35_cast_fp16")]; tensor input_37_pad_type_0 = const()[name = tensor("input_37_pad_type_0"), val = tensor("valid")]; tensor input_37_strides_0 = const()[name = tensor("input_37_strides_0"), val = tensor([1, 1])]; tensor input_37_pad_0 = const()[name = tensor("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_37_dilations_0 = const()[name = tensor("input_37_dilations_0"), val = tensor([1, 1])]; tensor input_37_groups_0 = const()[name = tensor("input_37_groups_0"), val = tensor(1)]; tensor layers_3_fc1_weight_to_fp16 = const()[name = tensor("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(317594688)))]; tensor layers_3_fc1_bias_to_fp16 = const()[name = tensor("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(330701952)))]; tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor("input_37_cast_fp16")]; tensor input_39_mode_0 = const()[name = tensor("input_39_mode_0"), val = tensor("EXACT")]; tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = tensor("input_39_cast_fp16")]; tensor hidden_states_9_pad_type_0 = const()[name = tensor("hidden_states_9_pad_type_0"), val = tensor("valid")]; tensor hidden_states_9_strides_0 = const()[name = tensor("hidden_states_9_strides_0"), val = tensor([1, 1])]; tensor hidden_states_9_pad_0 = const()[name = tensor("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_9_dilations_0 = const()[name = tensor("hidden_states_9_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_9_groups_0 = const()[name = tensor("hidden_states_9_groups_0"), val = tensor(1)]; tensor layers_3_fc2_weight_to_fp16 = const()[name = tensor("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(330712256)))]; tensor layers_3_fc2_bias_to_fp16 = const()[name = tensor("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343819520)))]; tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = tensor("hidden_states_9_cast_fp16")]; tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = tensor("inputs_25_cast_fp16")]; tensor var_1052 = const()[name = tensor("op_1052"), val = tensor(3)]; tensor out_25_axes_0 = const()[name = tensor("out_25_axes_0"), val = tensor([1])]; tensor var_1077_to_fp16 = const()[name = tensor("op_1077_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_1077_to_fp16, x = inputs_25_cast_fp16)[name = tensor("out_25_cast_fp16")]; tensor obj_57_gamma_0_to_fp16 = const()[name = tensor("obj_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343822144)))]; tensor obj_57_beta_0_to_fp16 = const()[name = tensor("obj_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343824768)))]; tensor obj_57_epsilon_0_to_fp16 = const()[name = tensor("obj_57_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = tensor("obj_57_cast_fp16")]; tensor query_17_pad_type_0 = const()[name = tensor("query_17_pad_type_0"), val = tensor("valid")]; tensor query_17_strides_0 = const()[name = tensor("query_17_strides_0"), val = tensor([1, 1])]; tensor query_17_pad_0 = const()[name = tensor("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_17_dilations_0 = const()[name = tensor("query_17_dilations_0"), val = tensor([1, 1])]; tensor query_17_groups_0 = const()[name = tensor("query_17_groups_0"), val = tensor(1)]; tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343827392)))]; tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(347104256)))]; tensor query_17_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor("query_17_cast_fp16")]; tensor current_key_9_pad_type_0 = const()[name = tensor("current_key_9_pad_type_0"), val = tensor("valid")]; tensor current_key_9_strides_0 = const()[name = tensor("current_key_9_strides_0"), val = tensor([1, 1])]; tensor current_key_9_pad_0 = const()[name = tensor("current_key_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_9_dilations_0 = const()[name = tensor("current_key_9_dilations_0"), val = tensor([1, 1])]; tensor current_key_9_groups_0 = const()[name = tensor("current_key_9_groups_0"), val = tensor(1)]; tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(347106880)))]; tensor current_key_9_cast_fp16 = conv(dilations = current_key_9_dilations_0, groups = current_key_9_groups_0, pad = current_key_9_pad_0, pad_type = current_key_9_pad_type_0, strides = current_key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor("current_key_9_cast_fp16")]; tensor current_value_9_pad_type_0 = const()[name = tensor("current_value_9_pad_type_0"), val = tensor("valid")]; tensor current_value_9_strides_0 = const()[name = tensor("current_value_9_strides_0"), val = tensor([1, 1])]; tensor current_value_9_pad_0 = const()[name = tensor("current_value_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_9_dilations_0 = const()[name = tensor("current_value_9_dilations_0"), val = tensor([1, 1])]; tensor current_value_9_groups_0 = const()[name = tensor("current_value_9_groups_0"), val = tensor(1)]; tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350383744)))]; tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353660608)))]; tensor current_value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = current_value_9_dilations_0, groups = current_value_9_groups_0, pad = current_value_9_pad_0, pad_type = current_value_9_pad_type_0, strides = current_value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor("current_value_9_cast_fp16")]; tensor var_1116_cast_fp16 = mul(x = var_103_cast_fp16_4, y = var_239_cast_fp16)[name = tensor("op_1116_cast_fp16")]; tensor var_1117_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1117_cast_fp16")]; tensor key_17_cast_fp16 = add(x = var_1116_cast_fp16, y = var_1117_cast_fp16)[name = tensor("key_17_cast_fp16")]; tensor var_1120_cast_fp16 = mul(x = var_138_cast_fp16_4, y = var_239_cast_fp16)[name = tensor("op_1120_cast_fp16")]; tensor var_1121_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1121_cast_fp16")]; tensor value_17_cast_fp16 = add(x = var_1120_cast_fp16, y = var_1121_cast_fp16)[name = tensor("value_17_cast_fp16")]; tensor var_1125 = const()[name = tensor("op_1125"), val = tensor([1, 20, 64, 1])]; tensor mh_q_17_cast_fp16 = reshape(shape = var_1125, x = query_17_cast_fp16)[name = tensor("mh_q_17_cast_fp16")]; tensor var_1127_to_fp16 = const()[name = tensor("op_1127_to_fp16"), val = tensor(0x1p-3)]; tensor var_1128_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1127_to_fp16)[name = tensor("op_1128_cast_fp16")]; tensor var_1131 = const()[name = tensor("op_1131"), val = tensor([1, 20, 64, 448])]; tensor var_1132_cast_fp16 = reshape(shape = var_1131, x = key_17_cast_fp16)[name = tensor("op_1132_cast_fp16")]; tensor mh_w_25_transpose_x_0 = const()[name = tensor("mh_w_25_transpose_x_0"), val = tensor(true)]; tensor mh_w_25_transpose_y_0 = const()[name = tensor("mh_w_25_transpose_y_0"), val = tensor(false)]; tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_1128_cast_fp16, y = var_1132_cast_fp16)[name = tensor("mh_w_25_cast_fp16")]; tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_27_cast_fp16")]; tensor var_1140_cast_fp16 = softmax(axis = var_1052, x = mh_w_27_cast_fp16)[name = tensor("op_1140_cast_fp16")]; tensor var_1141 = const()[name = tensor("op_1141"), val = tensor([1, 20, 64, 448])]; tensor var_1142_cast_fp16 = reshape(shape = var_1141, x = value_17_cast_fp16)[name = tensor("op_1142_cast_fp16")]; tensor attn_17_transpose_x_0 = const()[name = tensor("attn_17_transpose_x_0"), val = tensor(false)]; tensor attn_17_transpose_y_0 = const()[name = tensor("attn_17_transpose_y_0"), val = tensor(true)]; tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1142_cast_fp16, y = var_1140_cast_fp16)[name = tensor("attn_17_cast_fp16")]; tensor var_1145 = const()[name = tensor("op_1145"), val = tensor([1, 1280, 1, 1])]; tensor input_41_cast_fp16 = reshape(shape = var_1145, x = attn_17_cast_fp16)[name = tensor("input_41_cast_fp16")]; tensor obj_63_pad_type_0 = const()[name = tensor("obj_63_pad_type_0"), val = tensor("valid")]; tensor obj_63_strides_0 = const()[name = tensor("obj_63_strides_0"), val = tensor([1, 1])]; tensor obj_63_pad_0 = const()[name = tensor("obj_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_63_dilations_0 = const()[name = tensor("obj_63_dilations_0"), val = tensor([1, 1])]; tensor obj_63_groups_0 = const()[name = tensor("obj_63_groups_0"), val = tensor(1)]; tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353663232)))]; tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(356940096)))]; tensor obj_63_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_63_dilations_0, groups = obj_63_groups_0, pad = obj_63_pad_0, pad_type = obj_63_pad_type_0, strides = obj_63_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = tensor("obj_63_cast_fp16")]; tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_63_cast_fp16)[name = tensor("inputs_27_cast_fp16")]; tensor out_27_axes_0 = const()[name = tensor("out_27_axes_0"), val = tensor([1])]; tensor var_1167_to_fp16 = const()[name = tensor("op_1167_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1167_to_fp16, x = inputs_27_cast_fp16)[name = tensor("out_27_cast_fp16")]; tensor obj_65_gamma_0_to_fp16 = const()[name = tensor("obj_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(356942720)))]; tensor obj_65_beta_0_to_fp16 = const()[name = tensor("obj_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(356945344)))]; tensor obj_65_epsilon_0_to_fp16 = const()[name = tensor("obj_65_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = tensor("obj_65_cast_fp16")]; tensor query_19_pad_type_0 = const()[name = tensor("query_19_pad_type_0"), val = tensor("valid")]; tensor query_19_strides_0 = const()[name = tensor("query_19_strides_0"), val = tensor([1, 1])]; tensor query_19_pad_0 = const()[name = tensor("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_19_dilations_0 = const()[name = tensor("query_19_dilations_0"), val = tensor([1, 1])]; tensor query_19_groups_0 = const()[name = tensor("query_19_groups_0"), val = tensor(1)]; tensor layers_4_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_4_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(356947968)))]; tensor layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360224832)))]; tensor query_19_cast_fp16 = conv(bias = layers_4_encoder_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_4_encoder_attn_q_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = tensor("query_19_cast_fp16")]; tensor key_19_pad_type_0 = const()[name = tensor("key_19_pad_type_0"), val = tensor("valid")]; tensor key_19_strides_0 = const()[name = tensor("key_19_strides_0"), val = tensor([1, 1])]; tensor key_19_pad_0 = const()[name = tensor("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_19_dilations_0 = const()[name = tensor("key_19_dilations_0"), val = tensor([1, 1])]; tensor key_19_groups_0 = const()[name = tensor("key_19_groups_0"), val = tensor(1)]; tensor layers_4_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_4_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360227456)))]; tensor key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_4_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_19_cast_fp16")]; tensor value_19_pad_type_0 = const()[name = tensor("value_19_pad_type_0"), val = tensor("valid")]; tensor value_19_strides_0 = const()[name = tensor("value_19_strides_0"), val = tensor([1, 1])]; tensor value_19_pad_0 = const()[name = tensor("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_19_dilations_0 = const()[name = tensor("value_19_dilations_0"), val = tensor([1, 1])]; tensor value_19_groups_0 = const()[name = tensor("value_19_groups_0"), val = tensor(1)]; tensor layers_4_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_4_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(363504320)))]; tensor layers_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366781184)))]; tensor value_19_cast_fp16 = conv(bias = layers_4_encoder_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_4_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_19_cast_fp16")]; tensor var_1203 = const()[name = tensor("op_1203"), val = tensor([1, 20, 64, 1])]; tensor mh_q_19_cast_fp16 = reshape(shape = var_1203, x = query_19_cast_fp16)[name = tensor("mh_q_19_cast_fp16")]; tensor var_1205_to_fp16 = const()[name = tensor("op_1205_to_fp16"), val = tensor(0x1p-3)]; tensor var_1206_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1205_to_fp16)[name = tensor("op_1206_cast_fp16")]; tensor var_1209 = const()[name = tensor("op_1209"), val = tensor([1, 20, 64, 1500])]; tensor var_1210_cast_fp16 = reshape(shape = var_1209, x = key_19_cast_fp16)[name = tensor("op_1210_cast_fp16")]; tensor mh_w_29_transpose_x_0 = const()[name = tensor("mh_w_29_transpose_x_0"), val = tensor(true)]; tensor mh_w_29_transpose_y_0 = const()[name = tensor("mh_w_29_transpose_y_0"), val = tensor(false)]; tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_1206_cast_fp16, y = var_1210_cast_fp16)[name = tensor("mh_w_29_cast_fp16")]; tensor obj_69_cast_fp16 = softmax(axis = var_1052, x = mh_w_29_cast_fp16)[name = tensor("obj_69_cast_fp16")]; tensor var_1214 = const()[name = tensor("op_1214"), val = tensor([1, 20, 64, 1500])]; tensor var_1215_cast_fp16 = reshape(shape = var_1214, x = value_19_cast_fp16)[name = tensor("op_1215_cast_fp16")]; tensor attn_19_transpose_x_0 = const()[name = tensor("attn_19_transpose_x_0"), val = tensor(false)]; tensor attn_19_transpose_y_0 = const()[name = tensor("attn_19_transpose_y_0"), val = tensor(true)]; tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1215_cast_fp16, y = obj_69_cast_fp16)[name = tensor("attn_19_cast_fp16")]; tensor var_1218 = const()[name = tensor("op_1218"), val = tensor([1, 1280, 1, 1])]; tensor input_43_cast_fp16 = reshape(shape = var_1218, x = attn_19_cast_fp16)[name = tensor("input_43_cast_fp16")]; tensor obj_67_pad_type_0 = const()[name = tensor("obj_67_pad_type_0"), val = tensor("valid")]; tensor obj_67_strides_0 = const()[name = tensor("obj_67_strides_0"), val = tensor([1, 1])]; tensor obj_67_pad_0 = const()[name = tensor("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_67_dilations_0 = const()[name = tensor("obj_67_dilations_0"), val = tensor([1, 1])]; tensor obj_67_groups_0 = const()[name = tensor("obj_67_groups_0"), val = tensor(1)]; tensor layers_4_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_4_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366783808)))]; tensor layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370060672)))]; tensor obj_67_cast_fp16 = conv(bias = layers_4_encoder_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_4_encoder_attn_o_proj_weight_to_fp16, x = input_43_cast_fp16)[name = tensor("obj_67_cast_fp16")]; tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_67_cast_fp16)[name = tensor("inputs_29_cast_fp16")]; tensor out_29_axes_0 = const()[name = tensor("out_29_axes_0"), val = tensor([1])]; tensor var_1236_to_fp16 = const()[name = tensor("op_1236_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1236_to_fp16, x = inputs_29_cast_fp16)[name = tensor("out_29_cast_fp16")]; tensor input_45_gamma_0_to_fp16 = const()[name = tensor("input_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370063296)))]; tensor input_45_beta_0_to_fp16 = const()[name = tensor("input_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370065920)))]; tensor input_45_epsilon_0_to_fp16 = const()[name = tensor("input_45_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = tensor("input_45_cast_fp16")]; tensor input_47_pad_type_0 = const()[name = tensor("input_47_pad_type_0"), val = tensor("valid")]; tensor input_47_strides_0 = const()[name = tensor("input_47_strides_0"), val = tensor([1, 1])]; tensor input_47_pad_0 = const()[name = tensor("input_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_47_dilations_0 = const()[name = tensor("input_47_dilations_0"), val = tensor([1, 1])]; tensor input_47_groups_0 = const()[name = tensor("input_47_groups_0"), val = tensor(1)]; tensor layers_4_fc1_weight_to_fp16 = const()[name = tensor("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370068544)))]; tensor layers_4_fc1_bias_to_fp16 = const()[name = tensor("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383175808)))]; tensor input_47_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = tensor("input_47_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 = input_47_cast_fp16)[name = tensor("input_49_cast_fp16")]; tensor hidden_states_11_pad_type_0 = const()[name = tensor("hidden_states_11_pad_type_0"), val = tensor("valid")]; tensor hidden_states_11_strides_0 = const()[name = tensor("hidden_states_11_strides_0"), val = tensor([1, 1])]; tensor hidden_states_11_pad_0 = const()[name = tensor("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_11_dilations_0 = const()[name = tensor("hidden_states_11_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_11_groups_0 = const()[name = tensor("hidden_states_11_groups_0"), val = tensor(1)]; tensor layers_4_fc2_weight_to_fp16 = const()[name = tensor("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383186112)))]; tensor layers_4_fc2_bias_to_fp16 = const()[name = tensor("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396293376)))]; tensor hidden_states_11_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_49_cast_fp16)[name = tensor("hidden_states_11_cast_fp16")]; tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor("inputs_31_cast_fp16")]; tensor var_1271 = const()[name = tensor("op_1271"), val = tensor(3)]; tensor out_31_axes_0 = const()[name = tensor("out_31_axes_0"), val = tensor([1])]; tensor var_1296_to_fp16 = const()[name = tensor("op_1296_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1296_to_fp16, x = inputs_31_cast_fp16)[name = tensor("out_31_cast_fp16")]; tensor obj_71_gamma_0_to_fp16 = const()[name = tensor("obj_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396296000)))]; tensor obj_71_beta_0_to_fp16 = const()[name = tensor("obj_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396298624)))]; tensor obj_71_epsilon_0_to_fp16 = const()[name = tensor("obj_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_71_cast_fp16 = batch_norm(beta = obj_71_beta_0_to_fp16, epsilon = obj_71_epsilon_0_to_fp16, gamma = obj_71_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = tensor("obj_71_cast_fp16")]; tensor query_21_pad_type_0 = const()[name = tensor("query_21_pad_type_0"), val = tensor("valid")]; tensor query_21_strides_0 = const()[name = tensor("query_21_strides_0"), val = tensor([1, 1])]; tensor query_21_pad_0 = const()[name = tensor("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_21_dilations_0 = const()[name = tensor("query_21_dilations_0"), val = tensor([1, 1])]; tensor query_21_groups_0 = const()[name = tensor("query_21_groups_0"), val = tensor(1)]; tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396301248)))]; tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(399578112)))]; tensor query_21_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_71_cast_fp16)[name = tensor("query_21_cast_fp16")]; tensor current_key_11_pad_type_0 = const()[name = tensor("current_key_11_pad_type_0"), val = tensor("valid")]; tensor current_key_11_strides_0 = const()[name = tensor("current_key_11_strides_0"), val = tensor([1, 1])]; tensor current_key_11_pad_0 = const()[name = tensor("current_key_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_11_dilations_0 = const()[name = tensor("current_key_11_dilations_0"), val = tensor([1, 1])]; tensor current_key_11_groups_0 = const()[name = tensor("current_key_11_groups_0"), val = tensor(1)]; tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(399580736)))]; tensor current_key_11_cast_fp16 = conv(dilations = current_key_11_dilations_0, groups = current_key_11_groups_0, pad = current_key_11_pad_0, pad_type = current_key_11_pad_type_0, strides = current_key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_71_cast_fp16)[name = tensor("current_key_11_cast_fp16")]; tensor current_value_11_pad_type_0 = const()[name = tensor("current_value_11_pad_type_0"), val = tensor("valid")]; tensor current_value_11_strides_0 = const()[name = tensor("current_value_11_strides_0"), val = tensor([1, 1])]; tensor current_value_11_pad_0 = const()[name = tensor("current_value_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_11_dilations_0 = const()[name = tensor("current_value_11_dilations_0"), val = tensor([1, 1])]; tensor current_value_11_groups_0 = const()[name = tensor("current_value_11_groups_0"), val = tensor(1)]; tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(402857600)))]; tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406134464)))]; tensor current_value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = current_value_11_dilations_0, groups = current_value_11_groups_0, pad = current_value_11_pad_0, pad_type = current_value_11_pad_type_0, strides = current_value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_71_cast_fp16)[name = tensor("current_value_11_cast_fp16")]; tensor var_1335_cast_fp16 = mul(x = var_103_cast_fp16_5, y = var_239_cast_fp16)[name = tensor("op_1335_cast_fp16")]; tensor var_1336_cast_fp16 = mul(x = current_key_11_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1336_cast_fp16")]; tensor key_21_cast_fp16 = add(x = var_1335_cast_fp16, y = var_1336_cast_fp16)[name = tensor("key_21_cast_fp16")]; tensor var_1339_cast_fp16 = mul(x = var_138_cast_fp16_5, y = var_239_cast_fp16)[name = tensor("op_1339_cast_fp16")]; tensor var_1340_cast_fp16 = mul(x = current_value_11_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1340_cast_fp16")]; tensor value_21_cast_fp16 = add(x = var_1339_cast_fp16, y = var_1340_cast_fp16)[name = tensor("value_21_cast_fp16")]; tensor var_1344 = const()[name = tensor("op_1344"), val = tensor([1, 20, 64, 1])]; tensor mh_q_21_cast_fp16 = reshape(shape = var_1344, x = query_21_cast_fp16)[name = tensor("mh_q_21_cast_fp16")]; tensor var_1346_to_fp16 = const()[name = tensor("op_1346_to_fp16"), val = tensor(0x1p-3)]; tensor var_1347_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1346_to_fp16)[name = tensor("op_1347_cast_fp16")]; tensor var_1350 = const()[name = tensor("op_1350"), val = tensor([1, 20, 64, 448])]; tensor var_1351_cast_fp16 = reshape(shape = var_1350, x = key_21_cast_fp16)[name = tensor("op_1351_cast_fp16")]; tensor mh_w_31_transpose_x_0 = const()[name = tensor("mh_w_31_transpose_x_0"), val = tensor(true)]; tensor mh_w_31_transpose_y_0 = const()[name = tensor("mh_w_31_transpose_y_0"), val = tensor(false)]; tensor mh_w_31_cast_fp16 = matmul(transpose_x = mh_w_31_transpose_x_0, transpose_y = mh_w_31_transpose_y_0, x = var_1347_cast_fp16, y = var_1351_cast_fp16)[name = tensor("mh_w_31_cast_fp16")]; tensor mh_w_33_cast_fp16 = add(x = mh_w_31_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_33_cast_fp16")]; tensor var_1359_cast_fp16 = softmax(axis = var_1271, x = mh_w_33_cast_fp16)[name = tensor("op_1359_cast_fp16")]; tensor var_1360 = const()[name = tensor("op_1360"), val = tensor([1, 20, 64, 448])]; tensor var_1361_cast_fp16 = reshape(shape = var_1360, x = value_21_cast_fp16)[name = tensor("op_1361_cast_fp16")]; tensor attn_21_transpose_x_0 = const()[name = tensor("attn_21_transpose_x_0"), val = tensor(false)]; tensor attn_21_transpose_y_0 = const()[name = tensor("attn_21_transpose_y_0"), val = tensor(true)]; tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1361_cast_fp16, y = var_1359_cast_fp16)[name = tensor("attn_21_cast_fp16")]; tensor var_1364 = const()[name = tensor("op_1364"), val = tensor([1, 1280, 1, 1])]; tensor input_51_cast_fp16 = reshape(shape = var_1364, x = attn_21_cast_fp16)[name = tensor("input_51_cast_fp16")]; tensor obj_77_pad_type_0 = const()[name = tensor("obj_77_pad_type_0"), val = tensor("valid")]; tensor obj_77_strides_0 = const()[name = tensor("obj_77_strides_0"), val = tensor([1, 1])]; tensor obj_77_pad_0 = const()[name = tensor("obj_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_77_dilations_0 = const()[name = tensor("obj_77_dilations_0"), val = tensor([1, 1])]; tensor obj_77_groups_0 = const()[name = tensor("obj_77_groups_0"), val = tensor(1)]; tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406137088)))]; tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409413952)))]; tensor obj_77_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_77_dilations_0, groups = obj_77_groups_0, pad = obj_77_pad_0, pad_type = obj_77_pad_type_0, strides = obj_77_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_51_cast_fp16)[name = tensor("obj_77_cast_fp16")]; tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_77_cast_fp16)[name = tensor("inputs_33_cast_fp16")]; tensor out_33_axes_0 = const()[name = tensor("out_33_axes_0"), val = tensor([1])]; tensor var_1386_to_fp16 = const()[name = tensor("op_1386_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1386_to_fp16, x = inputs_33_cast_fp16)[name = tensor("out_33_cast_fp16")]; tensor obj_79_gamma_0_to_fp16 = const()[name = tensor("obj_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409416576)))]; tensor obj_79_beta_0_to_fp16 = const()[name = tensor("obj_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409419200)))]; tensor obj_79_epsilon_0_to_fp16 = const()[name = tensor("obj_79_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = tensor("obj_79_cast_fp16")]; tensor query_23_pad_type_0 = const()[name = tensor("query_23_pad_type_0"), val = tensor("valid")]; tensor query_23_strides_0 = const()[name = tensor("query_23_strides_0"), val = tensor([1, 1])]; tensor query_23_pad_0 = const()[name = tensor("query_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_23_dilations_0 = const()[name = tensor("query_23_dilations_0"), val = tensor([1, 1])]; tensor query_23_groups_0 = const()[name = tensor("query_23_groups_0"), val = tensor(1)]; tensor layers_5_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_5_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409421824)))]; tensor layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412698688)))]; tensor query_23_cast_fp16 = conv(bias = layers_5_encoder_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_5_encoder_attn_q_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = tensor("query_23_cast_fp16")]; tensor key_23_pad_type_0 = const()[name = tensor("key_23_pad_type_0"), val = tensor("valid")]; tensor key_23_strides_0 = const()[name = tensor("key_23_strides_0"), val = tensor([1, 1])]; tensor key_23_pad_0 = const()[name = tensor("key_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_23_dilations_0 = const()[name = tensor("key_23_dilations_0"), val = tensor([1, 1])]; tensor key_23_groups_0 = const()[name = tensor("key_23_groups_0"), val = tensor(1)]; tensor layers_5_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_5_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412701312)))]; tensor key_23_cast_fp16 = conv(dilations = key_23_dilations_0, groups = key_23_groups_0, pad = key_23_pad_0, pad_type = key_23_pad_type_0, strides = key_23_strides_0, weight = layers_5_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_23_cast_fp16")]; tensor value_23_pad_type_0 = const()[name = tensor("value_23_pad_type_0"), val = tensor("valid")]; tensor value_23_strides_0 = const()[name = tensor("value_23_strides_0"), val = tensor([1, 1])]; tensor value_23_pad_0 = const()[name = tensor("value_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_23_dilations_0 = const()[name = tensor("value_23_dilations_0"), val = tensor([1, 1])]; tensor value_23_groups_0 = const()[name = tensor("value_23_groups_0"), val = tensor(1)]; tensor layers_5_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_5_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415978176)))]; tensor layers_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419255040)))]; tensor value_23_cast_fp16 = conv(bias = layers_5_encoder_attn_v_proj_bias_to_fp16, dilations = value_23_dilations_0, groups = value_23_groups_0, pad = value_23_pad_0, pad_type = value_23_pad_type_0, strides = value_23_strides_0, weight = layers_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_23_cast_fp16")]; tensor var_1422 = const()[name = tensor("op_1422"), val = tensor([1, 20, 64, 1])]; tensor mh_q_23_cast_fp16 = reshape(shape = var_1422, x = query_23_cast_fp16)[name = tensor("mh_q_23_cast_fp16")]; tensor var_1424_to_fp16 = const()[name = tensor("op_1424_to_fp16"), val = tensor(0x1p-3)]; tensor var_1425_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1424_to_fp16)[name = tensor("op_1425_cast_fp16")]; tensor var_1428 = const()[name = tensor("op_1428"), val = tensor([1, 20, 64, 1500])]; tensor var_1429_cast_fp16 = reshape(shape = var_1428, x = key_23_cast_fp16)[name = tensor("op_1429_cast_fp16")]; tensor mh_w_35_transpose_x_0 = const()[name = tensor("mh_w_35_transpose_x_0"), val = tensor(true)]; tensor mh_w_35_transpose_y_0 = const()[name = tensor("mh_w_35_transpose_y_0"), val = tensor(false)]; tensor mh_w_35_cast_fp16 = matmul(transpose_x = mh_w_35_transpose_x_0, transpose_y = mh_w_35_transpose_y_0, x = var_1425_cast_fp16, y = var_1429_cast_fp16)[name = tensor("mh_w_35_cast_fp16")]; tensor obj_83_cast_fp16 = softmax(axis = var_1271, x = mh_w_35_cast_fp16)[name = tensor("obj_83_cast_fp16")]; tensor var_1433 = const()[name = tensor("op_1433"), val = tensor([1, 20, 64, 1500])]; tensor var_1434_cast_fp16 = reshape(shape = var_1433, x = value_23_cast_fp16)[name = tensor("op_1434_cast_fp16")]; tensor attn_23_transpose_x_0 = const()[name = tensor("attn_23_transpose_x_0"), val = tensor(false)]; tensor attn_23_transpose_y_0 = const()[name = tensor("attn_23_transpose_y_0"), val = tensor(true)]; tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1434_cast_fp16, y = obj_83_cast_fp16)[name = tensor("attn_23_cast_fp16")]; tensor var_1437 = const()[name = tensor("op_1437"), val = tensor([1, 1280, 1, 1])]; tensor input_53_cast_fp16 = reshape(shape = var_1437, x = attn_23_cast_fp16)[name = tensor("input_53_cast_fp16")]; tensor obj_81_pad_type_0 = const()[name = tensor("obj_81_pad_type_0"), val = tensor("valid")]; tensor obj_81_strides_0 = const()[name = tensor("obj_81_strides_0"), val = tensor([1, 1])]; tensor obj_81_pad_0 = const()[name = tensor("obj_81_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_81_dilations_0 = const()[name = tensor("obj_81_dilations_0"), val = tensor([1, 1])]; tensor obj_81_groups_0 = const()[name = tensor("obj_81_groups_0"), val = tensor(1)]; tensor layers_5_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_5_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419257664)))]; tensor layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422534528)))]; tensor obj_81_cast_fp16 = conv(bias = layers_5_encoder_attn_o_proj_bias_to_fp16, dilations = obj_81_dilations_0, groups = obj_81_groups_0, pad = obj_81_pad_0, pad_type = obj_81_pad_type_0, strides = obj_81_strides_0, weight = layers_5_encoder_attn_o_proj_weight_to_fp16, x = input_53_cast_fp16)[name = tensor("obj_81_cast_fp16")]; tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_81_cast_fp16)[name = tensor("inputs_35_cast_fp16")]; tensor out_35_axes_0 = const()[name = tensor("out_35_axes_0"), val = tensor([1])]; tensor var_1455_to_fp16 = const()[name = tensor("op_1455_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1455_to_fp16, x = inputs_35_cast_fp16)[name = tensor("out_35_cast_fp16")]; tensor input_55_gamma_0_to_fp16 = const()[name = tensor("input_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422537152)))]; tensor input_55_beta_0_to_fp16 = const()[name = tensor("input_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422539776)))]; tensor input_55_epsilon_0_to_fp16 = const()[name = tensor("input_55_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_55_cast_fp16 = batch_norm(beta = input_55_beta_0_to_fp16, epsilon = input_55_epsilon_0_to_fp16, gamma = input_55_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = tensor("input_55_cast_fp16")]; tensor input_57_pad_type_0 = const()[name = tensor("input_57_pad_type_0"), val = tensor("valid")]; tensor input_57_strides_0 = const()[name = tensor("input_57_strides_0"), val = tensor([1, 1])]; tensor input_57_pad_0 = const()[name = tensor("input_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_57_dilations_0 = const()[name = tensor("input_57_dilations_0"), val = tensor([1, 1])]; tensor input_57_groups_0 = const()[name = tensor("input_57_groups_0"), val = tensor(1)]; tensor layers_5_fc1_weight_to_fp16 = const()[name = tensor("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422542400)))]; tensor layers_5_fc1_bias_to_fp16 = const()[name = tensor("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(435649664)))]; tensor input_57_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_55_cast_fp16)[name = tensor("input_57_cast_fp16")]; tensor input_59_mode_0 = const()[name = tensor("input_59_mode_0"), val = tensor("EXACT")]; tensor input_59_cast_fp16 = gelu(mode = input_59_mode_0, x = input_57_cast_fp16)[name = tensor("input_59_cast_fp16")]; tensor hidden_states_13_pad_type_0 = const()[name = tensor("hidden_states_13_pad_type_0"), val = tensor("valid")]; tensor hidden_states_13_strides_0 = const()[name = tensor("hidden_states_13_strides_0"), val = tensor([1, 1])]; tensor hidden_states_13_pad_0 = const()[name = tensor("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_13_dilations_0 = const()[name = tensor("hidden_states_13_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_13_groups_0 = const()[name = tensor("hidden_states_13_groups_0"), val = tensor(1)]; tensor layers_5_fc2_weight_to_fp16 = const()[name = tensor("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(435659968)))]; tensor layers_5_fc2_bias_to_fp16 = const()[name = tensor("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(448767232)))]; tensor hidden_states_13_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_59_cast_fp16)[name = tensor("hidden_states_13_cast_fp16")]; tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = tensor("inputs_37_cast_fp16")]; tensor var_1490 = const()[name = tensor("op_1490"), val = tensor(3)]; tensor out_37_axes_0 = const()[name = tensor("out_37_axes_0"), val = tensor([1])]; tensor var_1515_to_fp16 = const()[name = tensor("op_1515_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1515_to_fp16, x = inputs_37_cast_fp16)[name = tensor("out_37_cast_fp16")]; tensor obj_85_gamma_0_to_fp16 = const()[name = tensor("obj_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(448769856)))]; tensor obj_85_beta_0_to_fp16 = const()[name = tensor("obj_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(448772480)))]; tensor obj_85_epsilon_0_to_fp16 = const()[name = tensor("obj_85_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = tensor("obj_85_cast_fp16")]; tensor query_25_pad_type_0 = const()[name = tensor("query_25_pad_type_0"), val = tensor("valid")]; tensor query_25_strides_0 = const()[name = tensor("query_25_strides_0"), val = tensor([1, 1])]; tensor query_25_pad_0 = const()[name = tensor("query_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_25_dilations_0 = const()[name = tensor("query_25_dilations_0"), val = tensor([1, 1])]; tensor query_25_groups_0 = const()[name = tensor("query_25_groups_0"), val = tensor(1)]; tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(448775104)))]; tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(452051968)))]; tensor query_25_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor("query_25_cast_fp16")]; tensor current_key_13_pad_type_0 = const()[name = tensor("current_key_13_pad_type_0"), val = tensor("valid")]; tensor current_key_13_strides_0 = const()[name = tensor("current_key_13_strides_0"), val = tensor([1, 1])]; tensor current_key_13_pad_0 = const()[name = tensor("current_key_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_13_dilations_0 = const()[name = tensor("current_key_13_dilations_0"), val = tensor([1, 1])]; tensor current_key_13_groups_0 = const()[name = tensor("current_key_13_groups_0"), val = tensor(1)]; tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(452054592)))]; tensor current_key_13_cast_fp16 = conv(dilations = current_key_13_dilations_0, groups = current_key_13_groups_0, pad = current_key_13_pad_0, pad_type = current_key_13_pad_type_0, strides = current_key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor("current_key_13_cast_fp16")]; tensor current_value_13_pad_type_0 = const()[name = tensor("current_value_13_pad_type_0"), val = tensor("valid")]; tensor current_value_13_strides_0 = const()[name = tensor("current_value_13_strides_0"), val = tensor([1, 1])]; tensor current_value_13_pad_0 = const()[name = tensor("current_value_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_13_dilations_0 = const()[name = tensor("current_value_13_dilations_0"), val = tensor([1, 1])]; tensor current_value_13_groups_0 = const()[name = tensor("current_value_13_groups_0"), val = tensor(1)]; tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(455331456)))]; tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(458608320)))]; tensor current_value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = current_value_13_dilations_0, groups = current_value_13_groups_0, pad = current_value_13_pad_0, pad_type = current_value_13_pad_type_0, strides = current_value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor("current_value_13_cast_fp16")]; tensor var_1554_cast_fp16 = mul(x = var_103_cast_fp16_6, y = var_239_cast_fp16)[name = tensor("op_1554_cast_fp16")]; tensor var_1555_cast_fp16 = mul(x = current_key_13_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1555_cast_fp16")]; tensor key_25_cast_fp16 = add(x = var_1554_cast_fp16, y = var_1555_cast_fp16)[name = tensor("key_25_cast_fp16")]; tensor var_1558_cast_fp16 = mul(x = var_138_cast_fp16_6, y = var_239_cast_fp16)[name = tensor("op_1558_cast_fp16")]; tensor var_1559_cast_fp16 = mul(x = current_value_13_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1559_cast_fp16")]; tensor value_25_cast_fp16 = add(x = var_1558_cast_fp16, y = var_1559_cast_fp16)[name = tensor("value_25_cast_fp16")]; tensor var_1563 = const()[name = tensor("op_1563"), val = tensor([1, 20, 64, 1])]; tensor mh_q_25_cast_fp16 = reshape(shape = var_1563, x = query_25_cast_fp16)[name = tensor("mh_q_25_cast_fp16")]; tensor var_1565_to_fp16 = const()[name = tensor("op_1565_to_fp16"), val = tensor(0x1p-3)]; tensor var_1566_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1565_to_fp16)[name = tensor("op_1566_cast_fp16")]; tensor var_1569 = const()[name = tensor("op_1569"), val = tensor([1, 20, 64, 448])]; tensor var_1570_cast_fp16 = reshape(shape = var_1569, x = key_25_cast_fp16)[name = tensor("op_1570_cast_fp16")]; tensor mh_w_37_transpose_x_0 = const()[name = tensor("mh_w_37_transpose_x_0"), val = tensor(true)]; tensor mh_w_37_transpose_y_0 = const()[name = tensor("mh_w_37_transpose_y_0"), val = tensor(false)]; tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_1566_cast_fp16, y = var_1570_cast_fp16)[name = tensor("mh_w_37_cast_fp16")]; tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_39_cast_fp16")]; tensor var_1578_cast_fp16 = softmax(axis = var_1490, x = mh_w_39_cast_fp16)[name = tensor("op_1578_cast_fp16")]; tensor var_1579 = const()[name = tensor("op_1579"), val = tensor([1, 20, 64, 448])]; tensor var_1580_cast_fp16 = reshape(shape = var_1579, x = value_25_cast_fp16)[name = tensor("op_1580_cast_fp16")]; tensor attn_25_transpose_x_0 = const()[name = tensor("attn_25_transpose_x_0"), val = tensor(false)]; tensor attn_25_transpose_y_0 = const()[name = tensor("attn_25_transpose_y_0"), val = tensor(true)]; tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1580_cast_fp16, y = var_1578_cast_fp16)[name = tensor("attn_25_cast_fp16")]; tensor var_1583 = const()[name = tensor("op_1583"), val = tensor([1, 1280, 1, 1])]; tensor input_61_cast_fp16 = reshape(shape = var_1583, x = attn_25_cast_fp16)[name = tensor("input_61_cast_fp16")]; tensor obj_91_pad_type_0 = const()[name = tensor("obj_91_pad_type_0"), val = tensor("valid")]; tensor obj_91_strides_0 = const()[name = tensor("obj_91_strides_0"), val = tensor([1, 1])]; tensor obj_91_pad_0 = const()[name = tensor("obj_91_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_91_dilations_0 = const()[name = tensor("obj_91_dilations_0"), val = tensor([1, 1])]; tensor obj_91_groups_0 = const()[name = tensor("obj_91_groups_0"), val = tensor(1)]; tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(458610944)))]; tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(461887808)))]; tensor obj_91_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_91_dilations_0, groups = obj_91_groups_0, pad = obj_91_pad_0, pad_type = obj_91_pad_type_0, strides = obj_91_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_61_cast_fp16)[name = tensor("obj_91_cast_fp16")]; tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_91_cast_fp16)[name = tensor("inputs_39_cast_fp16")]; tensor out_39_axes_0 = const()[name = tensor("out_39_axes_0"), val = tensor([1])]; tensor var_1605_to_fp16 = const()[name = tensor("op_1605_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1605_to_fp16, x = inputs_39_cast_fp16)[name = tensor("out_39_cast_fp16")]; tensor obj_93_gamma_0_to_fp16 = const()[name = tensor("obj_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(461890432)))]; tensor obj_93_beta_0_to_fp16 = const()[name = tensor("obj_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(461893056)))]; tensor obj_93_epsilon_0_to_fp16 = const()[name = tensor("obj_93_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = tensor("obj_93_cast_fp16")]; tensor query_27_pad_type_0 = const()[name = tensor("query_27_pad_type_0"), val = tensor("valid")]; tensor query_27_strides_0 = const()[name = tensor("query_27_strides_0"), val = tensor([1, 1])]; tensor query_27_pad_0 = const()[name = tensor("query_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_27_dilations_0 = const()[name = tensor("query_27_dilations_0"), val = tensor([1, 1])]; tensor query_27_groups_0 = const()[name = tensor("query_27_groups_0"), val = tensor(1)]; tensor layers_6_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_6_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(461895680)))]; tensor layers_6_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_6_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465172544)))]; tensor query_27_cast_fp16 = conv(bias = layers_6_encoder_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_6_encoder_attn_q_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = tensor("query_27_cast_fp16")]; tensor key_27_pad_type_0 = const()[name = tensor("key_27_pad_type_0"), val = tensor("valid")]; tensor key_27_strides_0 = const()[name = tensor("key_27_strides_0"), val = tensor([1, 1])]; tensor key_27_pad_0 = const()[name = tensor("key_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_27_dilations_0 = const()[name = tensor("key_27_dilations_0"), val = tensor([1, 1])]; tensor key_27_groups_0 = const()[name = tensor("key_27_groups_0"), val = tensor(1)]; tensor layers_6_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_6_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465175168)))]; tensor key_27_cast_fp16 = conv(dilations = key_27_dilations_0, groups = key_27_groups_0, pad = key_27_pad_0, pad_type = key_27_pad_type_0, strides = key_27_strides_0, weight = layers_6_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_27_cast_fp16")]; tensor value_27_pad_type_0 = const()[name = tensor("value_27_pad_type_0"), val = tensor("valid")]; tensor value_27_strides_0 = const()[name = tensor("value_27_strides_0"), val = tensor([1, 1])]; tensor value_27_pad_0 = const()[name = tensor("value_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_27_dilations_0 = const()[name = tensor("value_27_dilations_0"), val = tensor([1, 1])]; tensor value_27_groups_0 = const()[name = tensor("value_27_groups_0"), val = tensor(1)]; tensor layers_6_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_6_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(468452032)))]; tensor layers_6_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_6_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471728896)))]; tensor value_27_cast_fp16 = conv(bias = layers_6_encoder_attn_v_proj_bias_to_fp16, dilations = value_27_dilations_0, groups = value_27_groups_0, pad = value_27_pad_0, pad_type = value_27_pad_type_0, strides = value_27_strides_0, weight = layers_6_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_27_cast_fp16")]; tensor var_1641 = const()[name = tensor("op_1641"), val = tensor([1, 20, 64, 1])]; tensor mh_q_27_cast_fp16 = reshape(shape = var_1641, x = query_27_cast_fp16)[name = tensor("mh_q_27_cast_fp16")]; tensor var_1643_to_fp16 = const()[name = tensor("op_1643_to_fp16"), val = tensor(0x1p-3)]; tensor var_1644_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1643_to_fp16)[name = tensor("op_1644_cast_fp16")]; tensor var_1647 = const()[name = tensor("op_1647"), val = tensor([1, 20, 64, 1500])]; tensor var_1648_cast_fp16 = reshape(shape = var_1647, x = key_27_cast_fp16)[name = tensor("op_1648_cast_fp16")]; tensor mh_w_41_transpose_x_0 = const()[name = tensor("mh_w_41_transpose_x_0"), val = tensor(true)]; tensor mh_w_41_transpose_y_0 = const()[name = tensor("mh_w_41_transpose_y_0"), val = tensor(false)]; tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_1644_cast_fp16, y = var_1648_cast_fp16)[name = tensor("mh_w_41_cast_fp16")]; tensor obj_97_cast_fp16 = softmax(axis = var_1490, x = mh_w_41_cast_fp16)[name = tensor("obj_97_cast_fp16")]; tensor var_1652 = const()[name = tensor("op_1652"), val = tensor([1, 20, 64, 1500])]; tensor var_1653_cast_fp16 = reshape(shape = var_1652, x = value_27_cast_fp16)[name = tensor("op_1653_cast_fp16")]; tensor attn_27_transpose_x_0 = const()[name = tensor("attn_27_transpose_x_0"), val = tensor(false)]; tensor attn_27_transpose_y_0 = const()[name = tensor("attn_27_transpose_y_0"), val = tensor(true)]; tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1653_cast_fp16, y = obj_97_cast_fp16)[name = tensor("attn_27_cast_fp16")]; tensor var_1656 = const()[name = tensor("op_1656"), val = tensor([1, 1280, 1, 1])]; tensor input_63_cast_fp16 = reshape(shape = var_1656, x = attn_27_cast_fp16)[name = tensor("input_63_cast_fp16")]; tensor obj_95_pad_type_0 = const()[name = tensor("obj_95_pad_type_0"), val = tensor("valid")]; tensor obj_95_strides_0 = const()[name = tensor("obj_95_strides_0"), val = tensor([1, 1])]; tensor obj_95_pad_0 = const()[name = tensor("obj_95_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_95_dilations_0 = const()[name = tensor("obj_95_dilations_0"), val = tensor([1, 1])]; tensor obj_95_groups_0 = const()[name = tensor("obj_95_groups_0"), val = tensor(1)]; tensor layers_6_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_6_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471731520)))]; tensor layers_6_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_6_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475008384)))]; tensor obj_95_cast_fp16 = conv(bias = layers_6_encoder_attn_o_proj_bias_to_fp16, dilations = obj_95_dilations_0, groups = obj_95_groups_0, pad = obj_95_pad_0, pad_type = obj_95_pad_type_0, strides = obj_95_strides_0, weight = layers_6_encoder_attn_o_proj_weight_to_fp16, x = input_63_cast_fp16)[name = tensor("obj_95_cast_fp16")]; tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = obj_95_cast_fp16)[name = tensor("inputs_41_cast_fp16")]; tensor out_41_axes_0 = const()[name = tensor("out_41_axes_0"), val = tensor([1])]; tensor var_1674_to_fp16 = const()[name = tensor("op_1674_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1674_to_fp16, x = inputs_41_cast_fp16)[name = tensor("out_41_cast_fp16")]; tensor input_65_gamma_0_to_fp16 = const()[name = tensor("input_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475011008)))]; tensor input_65_beta_0_to_fp16 = const()[name = tensor("input_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475013632)))]; tensor input_65_epsilon_0_to_fp16 = const()[name = tensor("input_65_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_65_cast_fp16 = batch_norm(beta = input_65_beta_0_to_fp16, epsilon = input_65_epsilon_0_to_fp16, gamma = input_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = tensor("input_65_cast_fp16")]; tensor input_67_pad_type_0 = const()[name = tensor("input_67_pad_type_0"), val = tensor("valid")]; tensor input_67_strides_0 = const()[name = tensor("input_67_strides_0"), val = tensor([1, 1])]; tensor input_67_pad_0 = const()[name = tensor("input_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_67_dilations_0 = const()[name = tensor("input_67_dilations_0"), val = tensor([1, 1])]; tensor input_67_groups_0 = const()[name = tensor("input_67_groups_0"), val = tensor(1)]; tensor layers_6_fc1_weight_to_fp16 = const()[name = tensor("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475016256)))]; tensor layers_6_fc1_bias_to_fp16 = const()[name = tensor("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488123520)))]; tensor input_67_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_65_cast_fp16)[name = tensor("input_67_cast_fp16")]; tensor input_69_mode_0 = const()[name = tensor("input_69_mode_0"), val = tensor("EXACT")]; tensor input_69_cast_fp16 = gelu(mode = input_69_mode_0, x = input_67_cast_fp16)[name = tensor("input_69_cast_fp16")]; tensor hidden_states_15_pad_type_0 = const()[name = tensor("hidden_states_15_pad_type_0"), val = tensor("valid")]; tensor hidden_states_15_strides_0 = const()[name = tensor("hidden_states_15_strides_0"), val = tensor([1, 1])]; tensor hidden_states_15_pad_0 = const()[name = tensor("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_15_dilations_0 = const()[name = tensor("hidden_states_15_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_15_groups_0 = const()[name = tensor("hidden_states_15_groups_0"), val = tensor(1)]; tensor layers_6_fc2_weight_to_fp16 = const()[name = tensor("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488133824)))]; tensor layers_6_fc2_bias_to_fp16 = const()[name = tensor("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(501241088)))]; tensor hidden_states_15_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_69_cast_fp16)[name = tensor("hidden_states_15_cast_fp16")]; tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = hidden_states_15_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; tensor var_1709 = const()[name = tensor("op_1709"), val = tensor(3)]; tensor out_43_axes_0 = const()[name = tensor("out_43_axes_0"), val = tensor([1])]; tensor var_1734_to_fp16 = const()[name = tensor("op_1734_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1734_to_fp16, x = inputs_43_cast_fp16)[name = tensor("out_43_cast_fp16")]; tensor obj_99_gamma_0_to_fp16 = const()[name = tensor("obj_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(501243712)))]; tensor obj_99_beta_0_to_fp16 = const()[name = tensor("obj_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(501246336)))]; tensor obj_99_epsilon_0_to_fp16 = const()[name = tensor("obj_99_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_99_cast_fp16 = batch_norm(beta = obj_99_beta_0_to_fp16, epsilon = obj_99_epsilon_0_to_fp16, gamma = obj_99_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = tensor("obj_99_cast_fp16")]; tensor query_29_pad_type_0 = const()[name = tensor("query_29_pad_type_0"), val = tensor("valid")]; tensor query_29_strides_0 = const()[name = tensor("query_29_strides_0"), val = tensor([1, 1])]; tensor query_29_pad_0 = const()[name = tensor("query_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_29_dilations_0 = const()[name = tensor("query_29_dilations_0"), val = tensor([1, 1])]; tensor query_29_groups_0 = const()[name = tensor("query_29_groups_0"), val = tensor(1)]; tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(501248960)))]; tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(504525824)))]; tensor query_29_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_99_cast_fp16)[name = tensor("query_29_cast_fp16")]; tensor current_key_15_pad_type_0 = const()[name = tensor("current_key_15_pad_type_0"), val = tensor("valid")]; tensor current_key_15_strides_0 = const()[name = tensor("current_key_15_strides_0"), val = tensor([1, 1])]; tensor current_key_15_pad_0 = const()[name = tensor("current_key_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_15_dilations_0 = const()[name = tensor("current_key_15_dilations_0"), val = tensor([1, 1])]; tensor current_key_15_groups_0 = const()[name = tensor("current_key_15_groups_0"), val = tensor(1)]; tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(504528448)))]; tensor current_key_15_cast_fp16 = conv(dilations = current_key_15_dilations_0, groups = current_key_15_groups_0, pad = current_key_15_pad_0, pad_type = current_key_15_pad_type_0, strides = current_key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_99_cast_fp16)[name = tensor("current_key_15_cast_fp16")]; tensor current_value_15_pad_type_0 = const()[name = tensor("current_value_15_pad_type_0"), val = tensor("valid")]; tensor current_value_15_strides_0 = const()[name = tensor("current_value_15_strides_0"), val = tensor([1, 1])]; tensor current_value_15_pad_0 = const()[name = tensor("current_value_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_15_dilations_0 = const()[name = tensor("current_value_15_dilations_0"), val = tensor([1, 1])]; tensor current_value_15_groups_0 = const()[name = tensor("current_value_15_groups_0"), val = tensor(1)]; tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(507805312)))]; tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(511082176)))]; tensor current_value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = current_value_15_dilations_0, groups = current_value_15_groups_0, pad = current_value_15_pad_0, pad_type = current_value_15_pad_type_0, strides = current_value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_99_cast_fp16)[name = tensor("current_value_15_cast_fp16")]; tensor var_1773_cast_fp16 = mul(x = var_103_cast_fp16_7, y = var_239_cast_fp16)[name = tensor("op_1773_cast_fp16")]; tensor var_1774_cast_fp16 = mul(x = current_key_15_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1774_cast_fp16")]; tensor key_29_cast_fp16 = add(x = var_1773_cast_fp16, y = var_1774_cast_fp16)[name = tensor("key_29_cast_fp16")]; tensor var_1777_cast_fp16 = mul(x = var_138_cast_fp16_7, y = var_239_cast_fp16)[name = tensor("op_1777_cast_fp16")]; tensor var_1778_cast_fp16 = mul(x = current_value_15_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1778_cast_fp16")]; tensor value_29_cast_fp16 = add(x = var_1777_cast_fp16, y = var_1778_cast_fp16)[name = tensor("value_29_cast_fp16")]; tensor var_1782 = const()[name = tensor("op_1782"), val = tensor([1, 20, 64, 1])]; tensor mh_q_29_cast_fp16 = reshape(shape = var_1782, x = query_29_cast_fp16)[name = tensor("mh_q_29_cast_fp16")]; tensor var_1784_to_fp16 = const()[name = tensor("op_1784_to_fp16"), val = tensor(0x1p-3)]; tensor var_1785_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1784_to_fp16)[name = tensor("op_1785_cast_fp16")]; tensor var_1788 = const()[name = tensor("op_1788"), val = tensor([1, 20, 64, 448])]; tensor var_1789_cast_fp16 = reshape(shape = var_1788, x = key_29_cast_fp16)[name = tensor("op_1789_cast_fp16")]; tensor mh_w_43_transpose_x_0 = const()[name = tensor("mh_w_43_transpose_x_0"), val = tensor(true)]; tensor mh_w_43_transpose_y_0 = const()[name = tensor("mh_w_43_transpose_y_0"), val = tensor(false)]; tensor mh_w_43_cast_fp16 = matmul(transpose_x = mh_w_43_transpose_x_0, transpose_y = mh_w_43_transpose_y_0, x = var_1785_cast_fp16, y = var_1789_cast_fp16)[name = tensor("mh_w_43_cast_fp16")]; tensor mh_w_45_cast_fp16 = add(x = mh_w_43_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_45_cast_fp16")]; tensor var_1797_cast_fp16 = softmax(axis = var_1709, x = mh_w_45_cast_fp16)[name = tensor("op_1797_cast_fp16")]; tensor var_1798 = const()[name = tensor("op_1798"), val = tensor([1, 20, 64, 448])]; tensor var_1799_cast_fp16 = reshape(shape = var_1798, x = value_29_cast_fp16)[name = tensor("op_1799_cast_fp16")]; tensor attn_29_transpose_x_0 = const()[name = tensor("attn_29_transpose_x_0"), val = tensor(false)]; tensor attn_29_transpose_y_0 = const()[name = tensor("attn_29_transpose_y_0"), val = tensor(true)]; tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1799_cast_fp16, y = var_1797_cast_fp16)[name = tensor("attn_29_cast_fp16")]; tensor var_1802 = const()[name = tensor("op_1802"), val = tensor([1, 1280, 1, 1])]; tensor input_71_cast_fp16 = reshape(shape = var_1802, x = attn_29_cast_fp16)[name = tensor("input_71_cast_fp16")]; tensor obj_105_pad_type_0 = const()[name = tensor("obj_105_pad_type_0"), val = tensor("valid")]; tensor obj_105_strides_0 = const()[name = tensor("obj_105_strides_0"), val = tensor([1, 1])]; tensor obj_105_pad_0 = const()[name = tensor("obj_105_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_105_dilations_0 = const()[name = tensor("obj_105_dilations_0"), val = tensor([1, 1])]; tensor obj_105_groups_0 = const()[name = tensor("obj_105_groups_0"), val = tensor(1)]; tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(511084800)))]; tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(514361664)))]; tensor obj_105_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_105_dilations_0, groups = obj_105_groups_0, pad = obj_105_pad_0, pad_type = obj_105_pad_type_0, strides = obj_105_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_71_cast_fp16)[name = tensor("obj_105_cast_fp16")]; tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_105_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; tensor out_45_axes_0 = const()[name = tensor("out_45_axes_0"), val = tensor([1])]; tensor var_1824_to_fp16 = const()[name = tensor("op_1824_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1824_to_fp16, x = inputs_45_cast_fp16)[name = tensor("out_45_cast_fp16")]; tensor obj_107_gamma_0_to_fp16 = const()[name = tensor("obj_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(514364288)))]; tensor obj_107_beta_0_to_fp16 = const()[name = tensor("obj_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(514366912)))]; tensor obj_107_epsilon_0_to_fp16 = const()[name = tensor("obj_107_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_107_cast_fp16 = batch_norm(beta = obj_107_beta_0_to_fp16, epsilon = obj_107_epsilon_0_to_fp16, gamma = obj_107_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = tensor("obj_107_cast_fp16")]; tensor query_31_pad_type_0 = const()[name = tensor("query_31_pad_type_0"), val = tensor("valid")]; tensor query_31_strides_0 = const()[name = tensor("query_31_strides_0"), val = tensor([1, 1])]; tensor query_31_pad_0 = const()[name = tensor("query_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_31_dilations_0 = const()[name = tensor("query_31_dilations_0"), val = tensor([1, 1])]; tensor query_31_groups_0 = const()[name = tensor("query_31_groups_0"), val = tensor(1)]; tensor layers_7_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_7_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(514369536)))]; tensor layers_7_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_7_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517646400)))]; tensor query_31_cast_fp16 = conv(bias = layers_7_encoder_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_7_encoder_attn_q_proj_weight_to_fp16, x = obj_107_cast_fp16)[name = tensor("query_31_cast_fp16")]; tensor key_31_pad_type_0 = const()[name = tensor("key_31_pad_type_0"), val = tensor("valid")]; tensor key_31_strides_0 = const()[name = tensor("key_31_strides_0"), val = tensor([1, 1])]; tensor key_31_pad_0 = const()[name = tensor("key_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_31_dilations_0 = const()[name = tensor("key_31_dilations_0"), val = tensor([1, 1])]; tensor key_31_groups_0 = const()[name = tensor("key_31_groups_0"), val = tensor(1)]; tensor layers_7_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_7_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517649024)))]; tensor key_31_cast_fp16 = conv(dilations = key_31_dilations_0, groups = key_31_groups_0, pad = key_31_pad_0, pad_type = key_31_pad_type_0, strides = key_31_strides_0, weight = layers_7_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_31_cast_fp16")]; tensor value_31_pad_type_0 = const()[name = tensor("value_31_pad_type_0"), val = tensor("valid")]; tensor value_31_strides_0 = const()[name = tensor("value_31_strides_0"), val = tensor([1, 1])]; tensor value_31_pad_0 = const()[name = tensor("value_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_31_dilations_0 = const()[name = tensor("value_31_dilations_0"), val = tensor([1, 1])]; tensor value_31_groups_0 = const()[name = tensor("value_31_groups_0"), val = tensor(1)]; tensor layers_7_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_7_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520925888)))]; tensor layers_7_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_7_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(524202752)))]; tensor value_31_cast_fp16 = conv(bias = layers_7_encoder_attn_v_proj_bias_to_fp16, dilations = value_31_dilations_0, groups = value_31_groups_0, pad = value_31_pad_0, pad_type = value_31_pad_type_0, strides = value_31_strides_0, weight = layers_7_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_31_cast_fp16")]; tensor var_1860 = const()[name = tensor("op_1860"), val = tensor([1, 20, 64, 1])]; tensor mh_q_31_cast_fp16 = reshape(shape = var_1860, x = query_31_cast_fp16)[name = tensor("mh_q_31_cast_fp16")]; tensor var_1862_to_fp16 = const()[name = tensor("op_1862_to_fp16"), val = tensor(0x1p-3)]; tensor var_1863_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_1862_to_fp16)[name = tensor("op_1863_cast_fp16")]; tensor var_1866 = const()[name = tensor("op_1866"), val = tensor([1, 20, 64, 1500])]; tensor var_1867_cast_fp16 = reshape(shape = var_1866, x = key_31_cast_fp16)[name = tensor("op_1867_cast_fp16")]; tensor mh_w_47_transpose_x_0 = const()[name = tensor("mh_w_47_transpose_x_0"), val = tensor(true)]; tensor mh_w_47_transpose_y_0 = const()[name = tensor("mh_w_47_transpose_y_0"), val = tensor(false)]; tensor mh_w_47_cast_fp16 = matmul(transpose_x = mh_w_47_transpose_x_0, transpose_y = mh_w_47_transpose_y_0, x = var_1863_cast_fp16, y = var_1867_cast_fp16)[name = tensor("mh_w_47_cast_fp16")]; tensor obj_111_cast_fp16 = softmax(axis = var_1709, x = mh_w_47_cast_fp16)[name = tensor("obj_111_cast_fp16")]; tensor var_1871 = const()[name = tensor("op_1871"), val = tensor([1, 20, 64, 1500])]; tensor var_1872_cast_fp16 = reshape(shape = var_1871, x = value_31_cast_fp16)[name = tensor("op_1872_cast_fp16")]; tensor attn_31_transpose_x_0 = const()[name = tensor("attn_31_transpose_x_0"), val = tensor(false)]; tensor attn_31_transpose_y_0 = const()[name = tensor("attn_31_transpose_y_0"), val = tensor(true)]; tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_1872_cast_fp16, y = obj_111_cast_fp16)[name = tensor("attn_31_cast_fp16")]; tensor var_1875 = const()[name = tensor("op_1875"), val = tensor([1, 1280, 1, 1])]; tensor input_73_cast_fp16 = reshape(shape = var_1875, x = attn_31_cast_fp16)[name = tensor("input_73_cast_fp16")]; tensor obj_109_pad_type_0 = const()[name = tensor("obj_109_pad_type_0"), val = tensor("valid")]; tensor obj_109_strides_0 = const()[name = tensor("obj_109_strides_0"), val = tensor([1, 1])]; tensor obj_109_pad_0 = const()[name = tensor("obj_109_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_109_dilations_0 = const()[name = tensor("obj_109_dilations_0"), val = tensor([1, 1])]; tensor obj_109_groups_0 = const()[name = tensor("obj_109_groups_0"), val = tensor(1)]; tensor layers_7_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_7_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(524205376)))]; tensor layers_7_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_7_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(527482240)))]; tensor obj_109_cast_fp16 = conv(bias = layers_7_encoder_attn_o_proj_bias_to_fp16, dilations = obj_109_dilations_0, groups = obj_109_groups_0, pad = obj_109_pad_0, pad_type = obj_109_pad_type_0, strides = obj_109_strides_0, weight = layers_7_encoder_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = tensor("obj_109_cast_fp16")]; tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_109_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; tensor out_47_axes_0 = const()[name = tensor("out_47_axes_0"), val = tensor([1])]; tensor var_1893_to_fp16 = const()[name = tensor("op_1893_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1893_to_fp16, x = inputs_47_cast_fp16)[name = tensor("out_47_cast_fp16")]; tensor input_75_gamma_0_to_fp16 = const()[name = tensor("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(527484864)))]; tensor input_75_beta_0_to_fp16 = const()[name = tensor("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(527487488)))]; tensor input_75_epsilon_0_to_fp16 = const()[name = tensor("input_75_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = tensor("input_75_cast_fp16")]; tensor input_77_pad_type_0 = const()[name = tensor("input_77_pad_type_0"), val = tensor("valid")]; tensor input_77_strides_0 = const()[name = tensor("input_77_strides_0"), val = tensor([1, 1])]; tensor input_77_pad_0 = const()[name = tensor("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_77_dilations_0 = const()[name = tensor("input_77_dilations_0"), val = tensor([1, 1])]; tensor input_77_groups_0 = const()[name = tensor("input_77_groups_0"), val = tensor(1)]; tensor layers_7_fc1_weight_to_fp16 = const()[name = tensor("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(527490112)))]; tensor layers_7_fc1_bias_to_fp16 = const()[name = tensor("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(540597376)))]; tensor input_77_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = tensor("input_77_cast_fp16")]; tensor input_79_mode_0 = const()[name = tensor("input_79_mode_0"), val = tensor("EXACT")]; tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = tensor("input_79_cast_fp16")]; tensor hidden_states_17_pad_type_0 = const()[name = tensor("hidden_states_17_pad_type_0"), val = tensor("valid")]; tensor hidden_states_17_strides_0 = const()[name = tensor("hidden_states_17_strides_0"), val = tensor([1, 1])]; tensor hidden_states_17_pad_0 = const()[name = tensor("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_17_dilations_0 = const()[name = tensor("hidden_states_17_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_17_groups_0 = const()[name = tensor("hidden_states_17_groups_0"), val = tensor(1)]; tensor layers_7_fc2_weight_to_fp16 = const()[name = tensor("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(540607680)))]; tensor layers_7_fc2_bias_to_fp16 = const()[name = tensor("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(553714944)))]; tensor hidden_states_17_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = tensor("hidden_states_17_cast_fp16")]; tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_17_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; tensor var_1928 = const()[name = tensor("op_1928"), val = tensor(3)]; tensor out_49_axes_0 = const()[name = tensor("out_49_axes_0"), val = tensor([1])]; tensor var_1953_to_fp16 = const()[name = tensor("op_1953_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1953_to_fp16, x = inputs_49_cast_fp16)[name = tensor("out_49_cast_fp16")]; tensor obj_113_gamma_0_to_fp16 = const()[name = tensor("obj_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(553717568)))]; tensor obj_113_beta_0_to_fp16 = const()[name = tensor("obj_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(553720192)))]; tensor obj_113_epsilon_0_to_fp16 = const()[name = tensor("obj_113_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_113_cast_fp16 = batch_norm(beta = obj_113_beta_0_to_fp16, epsilon = obj_113_epsilon_0_to_fp16, gamma = obj_113_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = tensor("obj_113_cast_fp16")]; tensor query_33_pad_type_0 = const()[name = tensor("query_33_pad_type_0"), val = tensor("valid")]; tensor query_33_strides_0 = const()[name = tensor("query_33_strides_0"), val = tensor([1, 1])]; tensor query_33_pad_0 = const()[name = tensor("query_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_33_dilations_0 = const()[name = tensor("query_33_dilations_0"), val = tensor([1, 1])]; tensor query_33_groups_0 = const()[name = tensor("query_33_groups_0"), val = tensor(1)]; tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(553722816)))]; tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(556999680)))]; tensor query_33_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = tensor("query_33_cast_fp16")]; tensor current_key_17_pad_type_0 = const()[name = tensor("current_key_17_pad_type_0"), val = tensor("valid")]; tensor current_key_17_strides_0 = const()[name = tensor("current_key_17_strides_0"), val = tensor([1, 1])]; tensor current_key_17_pad_0 = const()[name = tensor("current_key_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_17_dilations_0 = const()[name = tensor("current_key_17_dilations_0"), val = tensor([1, 1])]; tensor current_key_17_groups_0 = const()[name = tensor("current_key_17_groups_0"), val = tensor(1)]; tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(557002304)))]; tensor current_key_17_cast_fp16 = conv(dilations = current_key_17_dilations_0, groups = current_key_17_groups_0, pad = current_key_17_pad_0, pad_type = current_key_17_pad_type_0, strides = current_key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = tensor("current_key_17_cast_fp16")]; tensor current_value_17_pad_type_0 = const()[name = tensor("current_value_17_pad_type_0"), val = tensor("valid")]; tensor current_value_17_strides_0 = const()[name = tensor("current_value_17_strides_0"), val = tensor([1, 1])]; tensor current_value_17_pad_0 = const()[name = tensor("current_value_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_17_dilations_0 = const()[name = tensor("current_value_17_dilations_0"), val = tensor([1, 1])]; tensor current_value_17_groups_0 = const()[name = tensor("current_value_17_groups_0"), val = tensor(1)]; tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(560279168)))]; tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563556032)))]; tensor current_value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = current_value_17_dilations_0, groups = current_value_17_groups_0, pad = current_value_17_pad_0, pad_type = current_value_17_pad_type_0, strides = current_value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = tensor("current_value_17_cast_fp16")]; tensor var_1992_cast_fp16 = mul(x = var_103_cast_fp16_8, y = var_239_cast_fp16)[name = tensor("op_1992_cast_fp16")]; tensor var_1993_cast_fp16 = mul(x = current_key_17_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1993_cast_fp16")]; tensor key_33_cast_fp16 = add(x = var_1992_cast_fp16, y = var_1993_cast_fp16)[name = tensor("key_33_cast_fp16")]; tensor var_1996_cast_fp16 = mul(x = var_138_cast_fp16_8, y = var_239_cast_fp16)[name = tensor("op_1996_cast_fp16")]; tensor var_1997_cast_fp16 = mul(x = current_value_17_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_1997_cast_fp16")]; tensor value_33_cast_fp16 = add(x = var_1996_cast_fp16, y = var_1997_cast_fp16)[name = tensor("value_33_cast_fp16")]; tensor var_2001 = const()[name = tensor("op_2001"), val = tensor([1, 20, 64, 1])]; tensor mh_q_33_cast_fp16 = reshape(shape = var_2001, x = query_33_cast_fp16)[name = tensor("mh_q_33_cast_fp16")]; tensor var_2003_to_fp16 = const()[name = tensor("op_2003_to_fp16"), val = tensor(0x1p-3)]; tensor var_2004_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_2003_to_fp16)[name = tensor("op_2004_cast_fp16")]; tensor var_2007 = const()[name = tensor("op_2007"), val = tensor([1, 20, 64, 448])]; tensor var_2008_cast_fp16 = reshape(shape = var_2007, x = key_33_cast_fp16)[name = tensor("op_2008_cast_fp16")]; tensor mh_w_49_transpose_x_0 = const()[name = tensor("mh_w_49_transpose_x_0"), val = tensor(true)]; tensor mh_w_49_transpose_y_0 = const()[name = tensor("mh_w_49_transpose_y_0"), val = tensor(false)]; tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_2004_cast_fp16, y = var_2008_cast_fp16)[name = tensor("mh_w_49_cast_fp16")]; tensor mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_51_cast_fp16")]; tensor var_2016_cast_fp16 = softmax(axis = var_1928, x = mh_w_51_cast_fp16)[name = tensor("op_2016_cast_fp16")]; tensor var_2017 = const()[name = tensor("op_2017"), val = tensor([1, 20, 64, 448])]; tensor var_2018_cast_fp16 = reshape(shape = var_2017, x = value_33_cast_fp16)[name = tensor("op_2018_cast_fp16")]; tensor attn_33_transpose_x_0 = const()[name = tensor("attn_33_transpose_x_0"), val = tensor(false)]; tensor attn_33_transpose_y_0 = const()[name = tensor("attn_33_transpose_y_0"), val = tensor(true)]; tensor attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_2018_cast_fp16, y = var_2016_cast_fp16)[name = tensor("attn_33_cast_fp16")]; tensor var_2021 = const()[name = tensor("op_2021"), val = tensor([1, 1280, 1, 1])]; tensor input_81_cast_fp16 = reshape(shape = var_2021, x = attn_33_cast_fp16)[name = tensor("input_81_cast_fp16")]; tensor obj_119_pad_type_0 = const()[name = tensor("obj_119_pad_type_0"), val = tensor("valid")]; tensor obj_119_strides_0 = const()[name = tensor("obj_119_strides_0"), val = tensor([1, 1])]; tensor obj_119_pad_0 = const()[name = tensor("obj_119_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_119_dilations_0 = const()[name = tensor("obj_119_dilations_0"), val = tensor([1, 1])]; tensor obj_119_groups_0 = const()[name = tensor("obj_119_groups_0"), val = tensor(1)]; tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563558656)))]; tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566835520)))]; tensor obj_119_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_119_dilations_0, groups = obj_119_groups_0, pad = obj_119_pad_0, pad_type = obj_119_pad_type_0, strides = obj_119_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = tensor("obj_119_cast_fp16")]; tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_119_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; tensor out_51_axes_0 = const()[name = tensor("out_51_axes_0"), val = tensor([1])]; tensor var_2043_to_fp16 = const()[name = tensor("op_2043_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_2043_to_fp16, x = inputs_51_cast_fp16)[name = tensor("out_51_cast_fp16")]; tensor obj_121_gamma_0_to_fp16 = const()[name = tensor("obj_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566838144)))]; tensor obj_121_beta_0_to_fp16 = const()[name = tensor("obj_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566840768)))]; tensor obj_121_epsilon_0_to_fp16 = const()[name = tensor("obj_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_121_cast_fp16 = batch_norm(beta = obj_121_beta_0_to_fp16, epsilon = obj_121_epsilon_0_to_fp16, gamma = obj_121_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = tensor("obj_121_cast_fp16")]; tensor query_35_pad_type_0 = const()[name = tensor("query_35_pad_type_0"), val = tensor("valid")]; tensor query_35_strides_0 = const()[name = tensor("query_35_strides_0"), val = tensor([1, 1])]; tensor query_35_pad_0 = const()[name = tensor("query_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_35_dilations_0 = const()[name = tensor("query_35_dilations_0"), val = tensor([1, 1])]; tensor query_35_groups_0 = const()[name = tensor("query_35_groups_0"), val = tensor(1)]; tensor layers_8_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_8_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566843392)))]; tensor layers_8_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_8_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570120256)))]; tensor query_35_cast_fp16 = conv(bias = layers_8_encoder_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_8_encoder_attn_q_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = tensor("query_35_cast_fp16")]; tensor key_35_pad_type_0 = const()[name = tensor("key_35_pad_type_0"), val = tensor("valid")]; tensor key_35_strides_0 = const()[name = tensor("key_35_strides_0"), val = tensor([1, 1])]; tensor key_35_pad_0 = const()[name = tensor("key_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_35_dilations_0 = const()[name = tensor("key_35_dilations_0"), val = tensor([1, 1])]; tensor key_35_groups_0 = const()[name = tensor("key_35_groups_0"), val = tensor(1)]; tensor layers_8_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_8_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570122880)))]; tensor key_35_cast_fp16 = conv(dilations = key_35_dilations_0, groups = key_35_groups_0, pad = key_35_pad_0, pad_type = key_35_pad_type_0, strides = key_35_strides_0, weight = layers_8_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_35_cast_fp16")]; tensor value_35_pad_type_0 = const()[name = tensor("value_35_pad_type_0"), val = tensor("valid")]; tensor value_35_strides_0 = const()[name = tensor("value_35_strides_0"), val = tensor([1, 1])]; tensor value_35_pad_0 = const()[name = tensor("value_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_35_dilations_0 = const()[name = tensor("value_35_dilations_0"), val = tensor([1, 1])]; tensor value_35_groups_0 = const()[name = tensor("value_35_groups_0"), val = tensor(1)]; tensor layers_8_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_8_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(573399744)))]; tensor layers_8_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_8_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(576676608)))]; tensor value_35_cast_fp16 = conv(bias = layers_8_encoder_attn_v_proj_bias_to_fp16, dilations = value_35_dilations_0, groups = value_35_groups_0, pad = value_35_pad_0, pad_type = value_35_pad_type_0, strides = value_35_strides_0, weight = layers_8_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_35_cast_fp16")]; tensor var_2079 = const()[name = tensor("op_2079"), val = tensor([1, 20, 64, 1])]; tensor mh_q_35_cast_fp16 = reshape(shape = var_2079, x = query_35_cast_fp16)[name = tensor("mh_q_35_cast_fp16")]; tensor var_2081_to_fp16 = const()[name = tensor("op_2081_to_fp16"), val = tensor(0x1p-3)]; tensor var_2082_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_2081_to_fp16)[name = tensor("op_2082_cast_fp16")]; tensor var_2085 = const()[name = tensor("op_2085"), val = tensor([1, 20, 64, 1500])]; tensor var_2086_cast_fp16 = reshape(shape = var_2085, x = key_35_cast_fp16)[name = tensor("op_2086_cast_fp16")]; tensor mh_w_53_transpose_x_0 = const()[name = tensor("mh_w_53_transpose_x_0"), val = tensor(true)]; tensor mh_w_53_transpose_y_0 = const()[name = tensor("mh_w_53_transpose_y_0"), val = tensor(false)]; tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_2082_cast_fp16, y = var_2086_cast_fp16)[name = tensor("mh_w_53_cast_fp16")]; tensor obj_125_cast_fp16 = softmax(axis = var_1928, x = mh_w_53_cast_fp16)[name = tensor("obj_125_cast_fp16")]; tensor var_2090 = const()[name = tensor("op_2090"), val = tensor([1, 20, 64, 1500])]; tensor var_2091_cast_fp16 = reshape(shape = var_2090, x = value_35_cast_fp16)[name = tensor("op_2091_cast_fp16")]; tensor attn_35_transpose_x_0 = const()[name = tensor("attn_35_transpose_x_0"), val = tensor(false)]; tensor attn_35_transpose_y_0 = const()[name = tensor("attn_35_transpose_y_0"), val = tensor(true)]; tensor attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_2091_cast_fp16, y = obj_125_cast_fp16)[name = tensor("attn_35_cast_fp16")]; tensor var_2094 = const()[name = tensor("op_2094"), val = tensor([1, 1280, 1, 1])]; tensor input_83_cast_fp16 = reshape(shape = var_2094, x = attn_35_cast_fp16)[name = tensor("input_83_cast_fp16")]; tensor obj_123_pad_type_0 = const()[name = tensor("obj_123_pad_type_0"), val = tensor("valid")]; tensor obj_123_strides_0 = const()[name = tensor("obj_123_strides_0"), val = tensor([1, 1])]; tensor obj_123_pad_0 = const()[name = tensor("obj_123_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_123_dilations_0 = const()[name = tensor("obj_123_dilations_0"), val = tensor([1, 1])]; tensor obj_123_groups_0 = const()[name = tensor("obj_123_groups_0"), val = tensor(1)]; tensor layers_8_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_8_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(576679232)))]; tensor layers_8_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_8_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(579956096)))]; tensor obj_123_cast_fp16 = conv(bias = layers_8_encoder_attn_o_proj_bias_to_fp16, dilations = obj_123_dilations_0, groups = obj_123_groups_0, pad = obj_123_pad_0, pad_type = obj_123_pad_type_0, strides = obj_123_strides_0, weight = layers_8_encoder_attn_o_proj_weight_to_fp16, x = input_83_cast_fp16)[name = tensor("obj_123_cast_fp16")]; tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = obj_123_cast_fp16)[name = tensor("inputs_53_cast_fp16")]; tensor out_53_axes_0 = const()[name = tensor("out_53_axes_0"), val = tensor([1])]; tensor var_2112_to_fp16 = const()[name = tensor("op_2112_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_2112_to_fp16, x = inputs_53_cast_fp16)[name = tensor("out_53_cast_fp16")]; tensor input_85_gamma_0_to_fp16 = const()[name = tensor("input_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(579958720)))]; tensor input_85_beta_0_to_fp16 = const()[name = tensor("input_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(579961344)))]; tensor input_85_epsilon_0_to_fp16 = const()[name = tensor("input_85_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_85_cast_fp16 = batch_norm(beta = input_85_beta_0_to_fp16, epsilon = input_85_epsilon_0_to_fp16, gamma = input_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = tensor("input_85_cast_fp16")]; tensor input_87_pad_type_0 = const()[name = tensor("input_87_pad_type_0"), val = tensor("valid")]; tensor input_87_strides_0 = const()[name = tensor("input_87_strides_0"), val = tensor([1, 1])]; tensor input_87_pad_0 = const()[name = tensor("input_87_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_87_dilations_0 = const()[name = tensor("input_87_dilations_0"), val = tensor([1, 1])]; tensor input_87_groups_0 = const()[name = tensor("input_87_groups_0"), val = tensor(1)]; tensor layers_8_fc1_weight_to_fp16 = const()[name = tensor("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(579963968)))]; tensor layers_8_fc1_bias_to_fp16 = const()[name = tensor("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593071232)))]; tensor input_87_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_87_dilations_0, groups = input_87_groups_0, pad = input_87_pad_0, pad_type = input_87_pad_type_0, strides = input_87_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_85_cast_fp16)[name = tensor("input_87_cast_fp16")]; tensor input_89_mode_0 = const()[name = tensor("input_89_mode_0"), val = tensor("EXACT")]; tensor input_89_cast_fp16 = gelu(mode = input_89_mode_0, x = input_87_cast_fp16)[name = tensor("input_89_cast_fp16")]; tensor hidden_states_19_pad_type_0 = const()[name = tensor("hidden_states_19_pad_type_0"), val = tensor("valid")]; tensor hidden_states_19_strides_0 = const()[name = tensor("hidden_states_19_strides_0"), val = tensor([1, 1])]; tensor hidden_states_19_pad_0 = const()[name = tensor("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_19_dilations_0 = const()[name = tensor("hidden_states_19_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_19_groups_0 = const()[name = tensor("hidden_states_19_groups_0"), val = tensor(1)]; tensor layers_8_fc2_weight_to_fp16 = const()[name = tensor("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593081536)))]; tensor layers_8_fc2_bias_to_fp16 = const()[name = tensor("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606188800)))]; tensor hidden_states_19_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_89_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = hidden_states_19_cast_fp16)[name = tensor("inputs_55_cast_fp16")]; tensor var_2147 = const()[name = tensor("op_2147"), val = tensor(3)]; tensor out_55_axes_0 = const()[name = tensor("out_55_axes_0"), val = tensor([1])]; tensor var_2172_to_fp16 = const()[name = tensor("op_2172_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2172_to_fp16, x = inputs_55_cast_fp16)[name = tensor("out_55_cast_fp16")]; tensor obj_127_gamma_0_to_fp16 = const()[name = tensor("obj_127_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606191424)))]; tensor obj_127_beta_0_to_fp16 = const()[name = tensor("obj_127_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606194048)))]; tensor obj_127_epsilon_0_to_fp16 = const()[name = tensor("obj_127_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_127_cast_fp16 = batch_norm(beta = obj_127_beta_0_to_fp16, epsilon = obj_127_epsilon_0_to_fp16, gamma = obj_127_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = tensor("obj_127_cast_fp16")]; tensor query_37_pad_type_0 = const()[name = tensor("query_37_pad_type_0"), val = tensor("valid")]; tensor query_37_strides_0 = const()[name = tensor("query_37_strides_0"), val = tensor([1, 1])]; tensor query_37_pad_0 = const()[name = tensor("query_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_37_dilations_0 = const()[name = tensor("query_37_dilations_0"), val = tensor([1, 1])]; tensor query_37_groups_0 = const()[name = tensor("query_37_groups_0"), val = tensor(1)]; tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606196672)))]; tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(609473536)))]; tensor query_37_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_127_cast_fp16)[name = tensor("query_37_cast_fp16")]; tensor current_key_19_pad_type_0 = const()[name = tensor("current_key_19_pad_type_0"), val = tensor("valid")]; tensor current_key_19_strides_0 = const()[name = tensor("current_key_19_strides_0"), val = tensor([1, 1])]; tensor current_key_19_pad_0 = const()[name = tensor("current_key_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_19_dilations_0 = const()[name = tensor("current_key_19_dilations_0"), val = tensor([1, 1])]; tensor current_key_19_groups_0 = const()[name = tensor("current_key_19_groups_0"), val = tensor(1)]; tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(609476160)))]; tensor current_key_19_cast_fp16 = conv(dilations = current_key_19_dilations_0, groups = current_key_19_groups_0, pad = current_key_19_pad_0, pad_type = current_key_19_pad_type_0, strides = current_key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_127_cast_fp16)[name = tensor("current_key_19_cast_fp16")]; tensor current_value_19_pad_type_0 = const()[name = tensor("current_value_19_pad_type_0"), val = tensor("valid")]; tensor current_value_19_strides_0 = const()[name = tensor("current_value_19_strides_0"), val = tensor([1, 1])]; tensor current_value_19_pad_0 = const()[name = tensor("current_value_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_19_dilations_0 = const()[name = tensor("current_value_19_dilations_0"), val = tensor([1, 1])]; tensor current_value_19_groups_0 = const()[name = tensor("current_value_19_groups_0"), val = tensor(1)]; tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(612753024)))]; tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(616029888)))]; tensor current_value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = current_value_19_dilations_0, groups = current_value_19_groups_0, pad = current_value_19_pad_0, pad_type = current_value_19_pad_type_0, strides = current_value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_127_cast_fp16)[name = tensor("current_value_19_cast_fp16")]; tensor var_2211_cast_fp16 = mul(x = var_103_cast_fp16_9, y = var_239_cast_fp16)[name = tensor("op_2211_cast_fp16")]; tensor var_2212_cast_fp16 = mul(x = current_key_19_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_2212_cast_fp16")]; tensor key_37_cast_fp16 = add(x = var_2211_cast_fp16, y = var_2212_cast_fp16)[name = tensor("key_37_cast_fp16")]; tensor var_2215_cast_fp16 = mul(x = var_138_cast_fp16_9, y = var_239_cast_fp16)[name = tensor("op_2215_cast_fp16")]; tensor var_2216_cast_fp16 = mul(x = current_value_19_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_2216_cast_fp16")]; tensor value_37_cast_fp16 = add(x = var_2215_cast_fp16, y = var_2216_cast_fp16)[name = tensor("value_37_cast_fp16")]; tensor var_2220 = const()[name = tensor("op_2220"), val = tensor([1, 20, 64, 1])]; tensor mh_q_37_cast_fp16 = reshape(shape = var_2220, x = query_37_cast_fp16)[name = tensor("mh_q_37_cast_fp16")]; tensor var_2222_to_fp16 = const()[name = tensor("op_2222_to_fp16"), val = tensor(0x1p-3)]; tensor var_2223_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2222_to_fp16)[name = tensor("op_2223_cast_fp16")]; tensor var_2226 = const()[name = tensor("op_2226"), val = tensor([1, 20, 64, 448])]; tensor var_2227_cast_fp16 = reshape(shape = var_2226, x = key_37_cast_fp16)[name = tensor("op_2227_cast_fp16")]; tensor mh_w_55_transpose_x_0 = const()[name = tensor("mh_w_55_transpose_x_0"), val = tensor(true)]; tensor mh_w_55_transpose_y_0 = const()[name = tensor("mh_w_55_transpose_y_0"), val = tensor(false)]; tensor mh_w_55_cast_fp16 = matmul(transpose_x = mh_w_55_transpose_x_0, transpose_y = mh_w_55_transpose_y_0, x = var_2223_cast_fp16, y = var_2227_cast_fp16)[name = tensor("mh_w_55_cast_fp16")]; tensor mh_w_57_cast_fp16 = add(x = mh_w_55_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_57_cast_fp16")]; tensor var_2235_cast_fp16 = softmax(axis = var_2147, x = mh_w_57_cast_fp16)[name = tensor("op_2235_cast_fp16")]; tensor var_2236 = const()[name = tensor("op_2236"), val = tensor([1, 20, 64, 448])]; tensor var_2237_cast_fp16 = reshape(shape = var_2236, x = value_37_cast_fp16)[name = tensor("op_2237_cast_fp16")]; tensor attn_37_transpose_x_0 = const()[name = tensor("attn_37_transpose_x_0"), val = tensor(false)]; tensor attn_37_transpose_y_0 = const()[name = tensor("attn_37_transpose_y_0"), val = tensor(true)]; tensor attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2237_cast_fp16, y = var_2235_cast_fp16)[name = tensor("attn_37_cast_fp16")]; tensor var_2240 = const()[name = tensor("op_2240"), val = tensor([1, 1280, 1, 1])]; tensor input_91_cast_fp16 = reshape(shape = var_2240, x = attn_37_cast_fp16)[name = tensor("input_91_cast_fp16")]; tensor obj_133_pad_type_0 = const()[name = tensor("obj_133_pad_type_0"), val = tensor("valid")]; tensor obj_133_strides_0 = const()[name = tensor("obj_133_strides_0"), val = tensor([1, 1])]; tensor obj_133_pad_0 = const()[name = tensor("obj_133_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_133_dilations_0 = const()[name = tensor("obj_133_dilations_0"), val = tensor([1, 1])]; tensor obj_133_groups_0 = const()[name = tensor("obj_133_groups_0"), val = tensor(1)]; tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(616032512)))]; tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(619309376)))]; tensor obj_133_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_133_dilations_0, groups = obj_133_groups_0, pad = obj_133_pad_0, pad_type = obj_133_pad_type_0, strides = obj_133_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_91_cast_fp16)[name = tensor("obj_133_cast_fp16")]; tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = obj_133_cast_fp16)[name = tensor("inputs_57_cast_fp16")]; tensor out_57_axes_0 = const()[name = tensor("out_57_axes_0"), val = tensor([1])]; tensor var_2262_to_fp16 = const()[name = tensor("op_2262_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2262_to_fp16, x = inputs_57_cast_fp16)[name = tensor("out_57_cast_fp16")]; tensor obj_135_gamma_0_to_fp16 = const()[name = tensor("obj_135_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(619312000)))]; tensor obj_135_beta_0_to_fp16 = const()[name = tensor("obj_135_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(619314624)))]; tensor obj_135_epsilon_0_to_fp16 = const()[name = tensor("obj_135_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_135_cast_fp16 = batch_norm(beta = obj_135_beta_0_to_fp16, epsilon = obj_135_epsilon_0_to_fp16, gamma = obj_135_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = tensor("obj_135_cast_fp16")]; tensor query_39_pad_type_0 = const()[name = tensor("query_39_pad_type_0"), val = tensor("valid")]; tensor query_39_strides_0 = const()[name = tensor("query_39_strides_0"), val = tensor([1, 1])]; tensor query_39_pad_0 = const()[name = tensor("query_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_39_dilations_0 = const()[name = tensor("query_39_dilations_0"), val = tensor([1, 1])]; tensor query_39_groups_0 = const()[name = tensor("query_39_groups_0"), val = tensor(1)]; tensor layers_9_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_9_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(619317248)))]; tensor layers_9_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_9_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622594112)))]; tensor query_39_cast_fp16 = conv(bias = layers_9_encoder_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_9_encoder_attn_q_proj_weight_to_fp16, x = obj_135_cast_fp16)[name = tensor("query_39_cast_fp16")]; tensor key_39_pad_type_0 = const()[name = tensor("key_39_pad_type_0"), val = tensor("valid")]; tensor key_39_strides_0 = const()[name = tensor("key_39_strides_0"), val = tensor([1, 1])]; tensor key_39_pad_0 = const()[name = tensor("key_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_39_dilations_0 = const()[name = tensor("key_39_dilations_0"), val = tensor([1, 1])]; tensor key_39_groups_0 = const()[name = tensor("key_39_groups_0"), val = tensor(1)]; tensor layers_9_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_9_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622596736)))]; tensor key_39_cast_fp16 = conv(dilations = key_39_dilations_0, groups = key_39_groups_0, pad = key_39_pad_0, pad_type = key_39_pad_type_0, strides = key_39_strides_0, weight = layers_9_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_39_cast_fp16")]; tensor value_39_pad_type_0 = const()[name = tensor("value_39_pad_type_0"), val = tensor("valid")]; tensor value_39_strides_0 = const()[name = tensor("value_39_strides_0"), val = tensor([1, 1])]; tensor value_39_pad_0 = const()[name = tensor("value_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_39_dilations_0 = const()[name = tensor("value_39_dilations_0"), val = tensor([1, 1])]; tensor value_39_groups_0 = const()[name = tensor("value_39_groups_0"), val = tensor(1)]; tensor layers_9_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_9_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625873600)))]; tensor layers_9_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_9_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(629150464)))]; tensor value_39_cast_fp16 = conv(bias = layers_9_encoder_attn_v_proj_bias_to_fp16, dilations = value_39_dilations_0, groups = value_39_groups_0, pad = value_39_pad_0, pad_type = value_39_pad_type_0, strides = value_39_strides_0, weight = layers_9_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_39_cast_fp16")]; tensor var_2298 = const()[name = tensor("op_2298"), val = tensor([1, 20, 64, 1])]; tensor mh_q_39_cast_fp16 = reshape(shape = var_2298, x = query_39_cast_fp16)[name = tensor("mh_q_39_cast_fp16")]; tensor var_2300_to_fp16 = const()[name = tensor("op_2300_to_fp16"), val = tensor(0x1p-3)]; tensor var_2301_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2300_to_fp16)[name = tensor("op_2301_cast_fp16")]; tensor var_2304 = const()[name = tensor("op_2304"), val = tensor([1, 20, 64, 1500])]; tensor var_2305_cast_fp16 = reshape(shape = var_2304, x = key_39_cast_fp16)[name = tensor("op_2305_cast_fp16")]; tensor mh_w_59_transpose_x_0 = const()[name = tensor("mh_w_59_transpose_x_0"), val = tensor(true)]; tensor mh_w_59_transpose_y_0 = const()[name = tensor("mh_w_59_transpose_y_0"), val = tensor(false)]; tensor mh_w_59_cast_fp16 = matmul(transpose_x = mh_w_59_transpose_x_0, transpose_y = mh_w_59_transpose_y_0, x = var_2301_cast_fp16, y = var_2305_cast_fp16)[name = tensor("mh_w_59_cast_fp16")]; tensor obj_139_cast_fp16 = softmax(axis = var_2147, x = mh_w_59_cast_fp16)[name = tensor("obj_139_cast_fp16")]; tensor var_2309 = const()[name = tensor("op_2309"), val = tensor([1, 20, 64, 1500])]; tensor var_2310_cast_fp16 = reshape(shape = var_2309, x = value_39_cast_fp16)[name = tensor("op_2310_cast_fp16")]; tensor attn_39_transpose_x_0 = const()[name = tensor("attn_39_transpose_x_0"), val = tensor(false)]; tensor attn_39_transpose_y_0 = const()[name = tensor("attn_39_transpose_y_0"), val = tensor(true)]; tensor attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2310_cast_fp16, y = obj_139_cast_fp16)[name = tensor("attn_39_cast_fp16")]; tensor var_2313 = const()[name = tensor("op_2313"), val = tensor([1, 1280, 1, 1])]; tensor input_93_cast_fp16 = reshape(shape = var_2313, x = attn_39_cast_fp16)[name = tensor("input_93_cast_fp16")]; tensor obj_137_pad_type_0 = const()[name = tensor("obj_137_pad_type_0"), val = tensor("valid")]; tensor obj_137_strides_0 = const()[name = tensor("obj_137_strides_0"), val = tensor([1, 1])]; tensor obj_137_pad_0 = const()[name = tensor("obj_137_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_137_dilations_0 = const()[name = tensor("obj_137_dilations_0"), val = tensor([1, 1])]; tensor obj_137_groups_0 = const()[name = tensor("obj_137_groups_0"), val = tensor(1)]; tensor layers_9_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_9_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(629153088)))]; tensor layers_9_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_9_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632429952)))]; tensor obj_137_cast_fp16 = conv(bias = layers_9_encoder_attn_o_proj_bias_to_fp16, dilations = obj_137_dilations_0, groups = obj_137_groups_0, pad = obj_137_pad_0, pad_type = obj_137_pad_type_0, strides = obj_137_strides_0, weight = layers_9_encoder_attn_o_proj_weight_to_fp16, x = input_93_cast_fp16)[name = tensor("obj_137_cast_fp16")]; tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_137_cast_fp16)[name = tensor("inputs_59_cast_fp16")]; tensor out_59_axes_0 = const()[name = tensor("out_59_axes_0"), val = tensor([1])]; tensor var_2331_to_fp16 = const()[name = tensor("op_2331_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2331_to_fp16, x = inputs_59_cast_fp16)[name = tensor("out_59_cast_fp16")]; tensor input_95_gamma_0_to_fp16 = const()[name = tensor("input_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632432576)))]; tensor input_95_beta_0_to_fp16 = const()[name = tensor("input_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632435200)))]; tensor input_95_epsilon_0_to_fp16 = const()[name = tensor("input_95_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = tensor("input_95_cast_fp16")]; tensor input_97_pad_type_0 = const()[name = tensor("input_97_pad_type_0"), val = tensor("valid")]; tensor input_97_strides_0 = const()[name = tensor("input_97_strides_0"), val = tensor([1, 1])]; tensor input_97_pad_0 = const()[name = tensor("input_97_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_97_dilations_0 = const()[name = tensor("input_97_dilations_0"), val = tensor([1, 1])]; tensor input_97_groups_0 = const()[name = tensor("input_97_groups_0"), val = tensor(1)]; tensor layers_9_fc1_weight_to_fp16 = const()[name = tensor("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632437824)))]; tensor layers_9_fc1_bias_to_fp16 = const()[name = tensor("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645545088)))]; tensor input_97_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_97_dilations_0, groups = input_97_groups_0, pad = input_97_pad_0, pad_type = input_97_pad_type_0, strides = input_97_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_95_cast_fp16)[name = tensor("input_97_cast_fp16")]; tensor input_99_mode_0 = const()[name = tensor("input_99_mode_0"), val = tensor("EXACT")]; tensor input_99_cast_fp16 = gelu(mode = input_99_mode_0, x = input_97_cast_fp16)[name = tensor("input_99_cast_fp16")]; tensor hidden_states_21_pad_type_0 = const()[name = tensor("hidden_states_21_pad_type_0"), val = tensor("valid")]; tensor hidden_states_21_strides_0 = const()[name = tensor("hidden_states_21_strides_0"), val = tensor([1, 1])]; tensor hidden_states_21_pad_0 = const()[name = tensor("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_21_dilations_0 = const()[name = tensor("hidden_states_21_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_21_groups_0 = const()[name = tensor("hidden_states_21_groups_0"), val = tensor(1)]; tensor layers_9_fc2_weight_to_fp16 = const()[name = tensor("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645555392)))]; tensor layers_9_fc2_bias_to_fp16 = const()[name = tensor("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(658662656)))]; tensor hidden_states_21_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_99_cast_fp16)[name = tensor("hidden_states_21_cast_fp16")]; tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_21_cast_fp16)[name = tensor("inputs_61_cast_fp16")]; tensor var_2366 = const()[name = tensor("op_2366"), val = tensor(3)]; tensor out_61_axes_0 = const()[name = tensor("out_61_axes_0"), val = tensor([1])]; tensor var_2391_to_fp16 = const()[name = tensor("op_2391_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2391_to_fp16, x = inputs_61_cast_fp16)[name = tensor("out_61_cast_fp16")]; tensor obj_141_gamma_0_to_fp16 = const()[name = tensor("obj_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(658665280)))]; tensor obj_141_beta_0_to_fp16 = const()[name = tensor("obj_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(658667904)))]; tensor obj_141_epsilon_0_to_fp16 = const()[name = tensor("obj_141_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_141_cast_fp16 = batch_norm(beta = obj_141_beta_0_to_fp16, epsilon = obj_141_epsilon_0_to_fp16, gamma = obj_141_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = tensor("obj_141_cast_fp16")]; tensor query_41_pad_type_0 = const()[name = tensor("query_41_pad_type_0"), val = tensor("valid")]; tensor query_41_strides_0 = const()[name = tensor("query_41_strides_0"), val = tensor([1, 1])]; tensor query_41_pad_0 = const()[name = tensor("query_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_41_dilations_0 = const()[name = tensor("query_41_dilations_0"), val = tensor([1, 1])]; tensor query_41_groups_0 = const()[name = tensor("query_41_groups_0"), val = tensor(1)]; tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(658670528)))]; tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(661947392)))]; tensor query_41_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_141_cast_fp16)[name = tensor("query_41_cast_fp16")]; tensor current_key_21_pad_type_0 = const()[name = tensor("current_key_21_pad_type_0"), val = tensor("valid")]; tensor current_key_21_strides_0 = const()[name = tensor("current_key_21_strides_0"), val = tensor([1, 1])]; tensor current_key_21_pad_0 = const()[name = tensor("current_key_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_21_dilations_0 = const()[name = tensor("current_key_21_dilations_0"), val = tensor([1, 1])]; tensor current_key_21_groups_0 = const()[name = tensor("current_key_21_groups_0"), val = tensor(1)]; tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(661950016)))]; tensor current_key_21_cast_fp16 = conv(dilations = current_key_21_dilations_0, groups = current_key_21_groups_0, pad = current_key_21_pad_0, pad_type = current_key_21_pad_type_0, strides = current_key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_141_cast_fp16)[name = tensor("current_key_21_cast_fp16")]; tensor current_value_21_pad_type_0 = const()[name = tensor("current_value_21_pad_type_0"), val = tensor("valid")]; tensor current_value_21_strides_0 = const()[name = tensor("current_value_21_strides_0"), val = tensor([1, 1])]; tensor current_value_21_pad_0 = const()[name = tensor("current_value_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_21_dilations_0 = const()[name = tensor("current_value_21_dilations_0"), val = tensor([1, 1])]; tensor current_value_21_groups_0 = const()[name = tensor("current_value_21_groups_0"), val = tensor(1)]; tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(665226880)))]; tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(668503744)))]; tensor current_value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = current_value_21_dilations_0, groups = current_value_21_groups_0, pad = current_value_21_pad_0, pad_type = current_value_21_pad_type_0, strides = current_value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_141_cast_fp16)[name = tensor("current_value_21_cast_fp16")]; tensor var_2430_cast_fp16 = mul(x = var_103_cast_fp16_10, y = var_239_cast_fp16)[name = tensor("op_2430_cast_fp16")]; tensor var_2431_cast_fp16 = mul(x = current_key_21_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_2431_cast_fp16")]; tensor key_41_cast_fp16 = add(x = var_2430_cast_fp16, y = var_2431_cast_fp16)[name = tensor("key_41_cast_fp16")]; tensor var_2434_cast_fp16 = mul(x = var_138_cast_fp16_10, y = var_239_cast_fp16)[name = tensor("op_2434_cast_fp16")]; tensor var_2435_cast_fp16 = mul(x = current_value_21_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_2435_cast_fp16")]; tensor value_41_cast_fp16 = add(x = var_2434_cast_fp16, y = var_2435_cast_fp16)[name = tensor("value_41_cast_fp16")]; tensor var_2439 = const()[name = tensor("op_2439"), val = tensor([1, 20, 64, 1])]; tensor mh_q_41_cast_fp16 = reshape(shape = var_2439, x = query_41_cast_fp16)[name = tensor("mh_q_41_cast_fp16")]; tensor var_2441_to_fp16 = const()[name = tensor("op_2441_to_fp16"), val = tensor(0x1p-3)]; tensor var_2442_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2441_to_fp16)[name = tensor("op_2442_cast_fp16")]; tensor var_2445 = const()[name = tensor("op_2445"), val = tensor([1, 20, 64, 448])]; tensor var_2446_cast_fp16 = reshape(shape = var_2445, x = key_41_cast_fp16)[name = tensor("op_2446_cast_fp16")]; tensor mh_w_61_transpose_x_0 = const()[name = tensor("mh_w_61_transpose_x_0"), val = tensor(true)]; tensor mh_w_61_transpose_y_0 = const()[name = tensor("mh_w_61_transpose_y_0"), val = tensor(false)]; tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_2442_cast_fp16, y = var_2446_cast_fp16)[name = tensor("mh_w_61_cast_fp16")]; tensor mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_63_cast_fp16")]; tensor var_2454_cast_fp16 = softmax(axis = var_2366, x = mh_w_63_cast_fp16)[name = tensor("op_2454_cast_fp16")]; tensor var_2455 = const()[name = tensor("op_2455"), val = tensor([1, 20, 64, 448])]; tensor var_2456_cast_fp16 = reshape(shape = var_2455, x = value_41_cast_fp16)[name = tensor("op_2456_cast_fp16")]; tensor attn_41_transpose_x_0 = const()[name = tensor("attn_41_transpose_x_0"), val = tensor(false)]; tensor attn_41_transpose_y_0 = const()[name = tensor("attn_41_transpose_y_0"), val = tensor(true)]; tensor attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2456_cast_fp16, y = var_2454_cast_fp16)[name = tensor("attn_41_cast_fp16")]; tensor var_2459 = const()[name = tensor("op_2459"), val = tensor([1, 1280, 1, 1])]; tensor input_101_cast_fp16 = reshape(shape = var_2459, x = attn_41_cast_fp16)[name = tensor("input_101_cast_fp16")]; tensor obj_147_pad_type_0 = const()[name = tensor("obj_147_pad_type_0"), val = tensor("valid")]; tensor obj_147_strides_0 = const()[name = tensor("obj_147_strides_0"), val = tensor([1, 1])]; tensor obj_147_pad_0 = const()[name = tensor("obj_147_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_147_dilations_0 = const()[name = tensor("obj_147_dilations_0"), val = tensor([1, 1])]; tensor obj_147_groups_0 = const()[name = tensor("obj_147_groups_0"), val = tensor(1)]; tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(668506368)))]; tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(671783232)))]; tensor obj_147_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_147_dilations_0, groups = obj_147_groups_0, pad = obj_147_pad_0, pad_type = obj_147_pad_type_0, strides = obj_147_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_101_cast_fp16)[name = tensor("obj_147_cast_fp16")]; tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_147_cast_fp16)[name = tensor("inputs_63_cast_fp16")]; tensor out_63_axes_0 = const()[name = tensor("out_63_axes_0"), val = tensor([1])]; tensor var_2481_to_fp16 = const()[name = tensor("op_2481_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2481_to_fp16, x = inputs_63_cast_fp16)[name = tensor("out_63_cast_fp16")]; tensor obj_149_gamma_0_to_fp16 = const()[name = tensor("obj_149_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(671785856)))]; tensor obj_149_beta_0_to_fp16 = const()[name = tensor("obj_149_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(671788480)))]; tensor obj_149_epsilon_0_to_fp16 = const()[name = tensor("obj_149_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_149_cast_fp16 = batch_norm(beta = obj_149_beta_0_to_fp16, epsilon = obj_149_epsilon_0_to_fp16, gamma = obj_149_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = tensor("obj_149_cast_fp16")]; tensor query_43_pad_type_0 = const()[name = tensor("query_43_pad_type_0"), val = tensor("valid")]; tensor query_43_strides_0 = const()[name = tensor("query_43_strides_0"), val = tensor([1, 1])]; tensor query_43_pad_0 = const()[name = tensor("query_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_43_dilations_0 = const()[name = tensor("query_43_dilations_0"), val = tensor([1, 1])]; tensor query_43_groups_0 = const()[name = tensor("query_43_groups_0"), val = tensor(1)]; tensor layers_10_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_10_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(671791104)))]; tensor layers_10_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_10_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(675067968)))]; tensor query_43_cast_fp16 = conv(bias = layers_10_encoder_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_10_encoder_attn_q_proj_weight_to_fp16, x = obj_149_cast_fp16)[name = tensor("query_43_cast_fp16")]; tensor key_43_pad_type_0 = const()[name = tensor("key_43_pad_type_0"), val = tensor("valid")]; tensor key_43_strides_0 = const()[name = tensor("key_43_strides_0"), val = tensor([1, 1])]; tensor key_43_pad_0 = const()[name = tensor("key_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_43_dilations_0 = const()[name = tensor("key_43_dilations_0"), val = tensor([1, 1])]; tensor key_43_groups_0 = const()[name = tensor("key_43_groups_0"), val = tensor(1)]; tensor layers_10_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_10_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(675070592)))]; tensor key_43_cast_fp16 = conv(dilations = key_43_dilations_0, groups = key_43_groups_0, pad = key_43_pad_0, pad_type = key_43_pad_type_0, strides = key_43_strides_0, weight = layers_10_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_43_cast_fp16")]; tensor value_43_pad_type_0 = const()[name = tensor("value_43_pad_type_0"), val = tensor("valid")]; tensor value_43_strides_0 = const()[name = tensor("value_43_strides_0"), val = tensor([1, 1])]; tensor value_43_pad_0 = const()[name = tensor("value_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_43_dilations_0 = const()[name = tensor("value_43_dilations_0"), val = tensor([1, 1])]; tensor value_43_groups_0 = const()[name = tensor("value_43_groups_0"), val = tensor(1)]; tensor layers_10_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_10_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(678347456)))]; tensor layers_10_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_10_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(681624320)))]; tensor value_43_cast_fp16 = conv(bias = layers_10_encoder_attn_v_proj_bias_to_fp16, dilations = value_43_dilations_0, groups = value_43_groups_0, pad = value_43_pad_0, pad_type = value_43_pad_type_0, strides = value_43_strides_0, weight = layers_10_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_43_cast_fp16")]; tensor var_2517 = const()[name = tensor("op_2517"), val = tensor([1, 20, 64, 1])]; tensor mh_q_43_cast_fp16 = reshape(shape = var_2517, x = query_43_cast_fp16)[name = tensor("mh_q_43_cast_fp16")]; tensor var_2519_to_fp16 = const()[name = tensor("op_2519_to_fp16"), val = tensor(0x1p-3)]; tensor var_2520_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2519_to_fp16)[name = tensor("op_2520_cast_fp16")]; tensor var_2523 = const()[name = tensor("op_2523"), val = tensor([1, 20, 64, 1500])]; tensor var_2524_cast_fp16 = reshape(shape = var_2523, x = key_43_cast_fp16)[name = tensor("op_2524_cast_fp16")]; tensor mh_w_65_transpose_x_0 = const()[name = tensor("mh_w_65_transpose_x_0"), val = tensor(true)]; tensor mh_w_65_transpose_y_0 = const()[name = tensor("mh_w_65_transpose_y_0"), val = tensor(false)]; tensor mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_2520_cast_fp16, y = var_2524_cast_fp16)[name = tensor("mh_w_65_cast_fp16")]; tensor obj_153_cast_fp16 = softmax(axis = var_2366, x = mh_w_65_cast_fp16)[name = tensor("obj_153_cast_fp16")]; tensor var_2528 = const()[name = tensor("op_2528"), val = tensor([1, 20, 64, 1500])]; tensor var_2529_cast_fp16 = reshape(shape = var_2528, x = value_43_cast_fp16)[name = tensor("op_2529_cast_fp16")]; tensor attn_43_transpose_x_0 = const()[name = tensor("attn_43_transpose_x_0"), val = tensor(false)]; tensor attn_43_transpose_y_0 = const()[name = tensor("attn_43_transpose_y_0"), val = tensor(true)]; tensor attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2529_cast_fp16, y = obj_153_cast_fp16)[name = tensor("attn_43_cast_fp16")]; tensor var_2532 = const()[name = tensor("op_2532"), val = tensor([1, 1280, 1, 1])]; tensor input_103_cast_fp16 = reshape(shape = var_2532, x = attn_43_cast_fp16)[name = tensor("input_103_cast_fp16")]; tensor obj_151_pad_type_0 = const()[name = tensor("obj_151_pad_type_0"), val = tensor("valid")]; tensor obj_151_strides_0 = const()[name = tensor("obj_151_strides_0"), val = tensor([1, 1])]; tensor obj_151_pad_0 = const()[name = tensor("obj_151_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_151_dilations_0 = const()[name = tensor("obj_151_dilations_0"), val = tensor([1, 1])]; tensor obj_151_groups_0 = const()[name = tensor("obj_151_groups_0"), val = tensor(1)]; tensor layers_10_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_10_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(681626944)))]; tensor layers_10_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_10_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684903808)))]; tensor obj_151_cast_fp16 = conv(bias = layers_10_encoder_attn_o_proj_bias_to_fp16, dilations = obj_151_dilations_0, groups = obj_151_groups_0, pad = obj_151_pad_0, pad_type = obj_151_pad_type_0, strides = obj_151_strides_0, weight = layers_10_encoder_attn_o_proj_weight_to_fp16, x = input_103_cast_fp16)[name = tensor("obj_151_cast_fp16")]; tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_151_cast_fp16)[name = tensor("inputs_65_cast_fp16")]; tensor out_65_axes_0 = const()[name = tensor("out_65_axes_0"), val = tensor([1])]; tensor var_2553_to_fp16 = const()[name = tensor("op_2553_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2553_to_fp16, x = inputs_65_cast_fp16)[name = tensor("out_65_cast_fp16")]; tensor input_105_gamma_0_to_fp16 = const()[name = tensor("input_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684906432)))]; tensor input_105_beta_0_to_fp16 = const()[name = tensor("input_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684909056)))]; tensor input_105_epsilon_0_to_fp16 = const()[name = tensor("input_105_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_105_cast_fp16 = batch_norm(beta = input_105_beta_0_to_fp16, epsilon = input_105_epsilon_0_to_fp16, gamma = input_105_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = tensor("input_105_cast_fp16")]; tensor input_107_pad_type_0 = const()[name = tensor("input_107_pad_type_0"), val = tensor("valid")]; tensor input_107_strides_0 = const()[name = tensor("input_107_strides_0"), val = tensor([1, 1])]; tensor input_107_pad_0 = const()[name = tensor("input_107_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_107_dilations_0 = const()[name = tensor("input_107_dilations_0"), val = tensor([1, 1])]; tensor input_107_groups_0 = const()[name = tensor("input_107_groups_0"), val = tensor(1)]; tensor layers_10_fc1_weight_to_fp16 = const()[name = tensor("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(684911680)))]; tensor layers_10_fc1_bias_to_fp16 = const()[name = tensor("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(698018944)))]; tensor input_107_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_107_dilations_0, groups = input_107_groups_0, pad = input_107_pad_0, pad_type = input_107_pad_type_0, strides = input_107_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_105_cast_fp16)[name = tensor("input_107_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 = input_107_cast_fp16)[name = tensor("input_109_cast_fp16")]; tensor hidden_states_23_pad_type_0 = const()[name = tensor("hidden_states_23_pad_type_0"), val = tensor("valid")]; tensor hidden_states_23_strides_0 = const()[name = tensor("hidden_states_23_strides_0"), val = tensor([1, 1])]; tensor hidden_states_23_pad_0 = const()[name = tensor("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_23_dilations_0 = const()[name = tensor("hidden_states_23_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_23_groups_0 = const()[name = tensor("hidden_states_23_groups_0"), val = tensor(1)]; tensor layers_10_fc2_weight_to_fp16 = const()[name = tensor("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(698029248)))]; tensor layers_10_fc2_bias_to_fp16 = const()[name = tensor("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711136512)))]; tensor hidden_states_23_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_109_cast_fp16)[name = tensor("hidden_states_23_cast_fp16")]; tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = hidden_states_23_cast_fp16)[name = tensor("inputs_67_cast_fp16")]; tensor var_2589 = const()[name = tensor("op_2589"), val = tensor(3)]; tensor out_67_axes_0 = const()[name = tensor("out_67_axes_0"), val = tensor([1])]; tensor var_2614_to_fp16 = const()[name = tensor("op_2614_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2614_to_fp16, x = inputs_67_cast_fp16)[name = tensor("out_67_cast_fp16")]; tensor obj_155_gamma_0_to_fp16 = const()[name = tensor("obj_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711139136)))]; tensor obj_155_beta_0_to_fp16 = const()[name = tensor("obj_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711141760)))]; tensor obj_155_epsilon_0_to_fp16 = const()[name = tensor("obj_155_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_155_cast_fp16 = batch_norm(beta = obj_155_beta_0_to_fp16, epsilon = obj_155_epsilon_0_to_fp16, gamma = obj_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = tensor("obj_155_cast_fp16")]; tensor query_45_pad_type_0 = const()[name = tensor("query_45_pad_type_0"), val = tensor("valid")]; tensor query_45_strides_0 = const()[name = tensor("query_45_strides_0"), val = tensor([1, 1])]; tensor query_45_pad_0 = const()[name = tensor("query_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_45_dilations_0 = const()[name = tensor("query_45_dilations_0"), val = tensor([1, 1])]; tensor query_45_groups_0 = const()[name = tensor("query_45_groups_0"), val = tensor(1)]; tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(711144384)))]; tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(714421248)))]; tensor query_45_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_155_cast_fp16)[name = tensor("query_45_cast_fp16")]; tensor current_key_23_pad_type_0 = const()[name = tensor("current_key_23_pad_type_0"), val = tensor("valid")]; tensor current_key_23_strides_0 = const()[name = tensor("current_key_23_strides_0"), val = tensor([1, 1])]; tensor current_key_23_pad_0 = const()[name = tensor("current_key_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_23_dilations_0 = const()[name = tensor("current_key_23_dilations_0"), val = tensor([1, 1])]; tensor current_key_23_groups_0 = const()[name = tensor("current_key_23_groups_0"), val = tensor(1)]; tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(714423872)))]; tensor current_key_23_cast_fp16 = conv(dilations = current_key_23_dilations_0, groups = current_key_23_groups_0, pad = current_key_23_pad_0, pad_type = current_key_23_pad_type_0, strides = current_key_23_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_155_cast_fp16)[name = tensor("current_key_23_cast_fp16")]; tensor current_value_23_pad_type_0 = const()[name = tensor("current_value_23_pad_type_0"), val = tensor("valid")]; tensor current_value_23_strides_0 = const()[name = tensor("current_value_23_strides_0"), val = tensor([1, 1])]; tensor current_value_23_pad_0 = const()[name = tensor("current_value_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_23_dilations_0 = const()[name = tensor("current_value_23_dilations_0"), val = tensor([1, 1])]; tensor current_value_23_groups_0 = const()[name = tensor("current_value_23_groups_0"), val = tensor(1)]; tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(717700736)))]; tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(720977600)))]; tensor current_value_23_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = current_value_23_dilations_0, groups = current_value_23_groups_0, pad = current_value_23_pad_0, pad_type = current_value_23_pad_type_0, strides = current_value_23_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_155_cast_fp16)[name = tensor("current_value_23_cast_fp16")]; tensor var_2653_cast_fp16 = mul(x = var_103_cast_fp16_11, y = var_239_cast_fp16)[name = tensor("op_2653_cast_fp16")]; tensor var_2654_cast_fp16 = mul(x = current_key_23_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_2654_cast_fp16")]; tensor key_45_cast_fp16 = add(x = var_2653_cast_fp16, y = var_2654_cast_fp16)[name = tensor("key_45_cast_fp16")]; tensor var_2657_cast_fp16 = mul(x = var_138_cast_fp16_11, y = var_239_cast_fp16)[name = tensor("op_2657_cast_fp16")]; tensor var_2658_cast_fp16 = mul(x = current_value_23_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_2658_cast_fp16")]; tensor value_45_cast_fp16 = add(x = var_2657_cast_fp16, y = var_2658_cast_fp16)[name = tensor("value_45_cast_fp16")]; tensor var_2662 = const()[name = tensor("op_2662"), val = tensor([1, 20, 64, 1])]; tensor mh_q_45_cast_fp16 = reshape(shape = var_2662, x = query_45_cast_fp16)[name = tensor("mh_q_45_cast_fp16")]; tensor var_2664_to_fp16 = const()[name = tensor("op_2664_to_fp16"), val = tensor(0x1p-3)]; tensor var_2665_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2664_to_fp16)[name = tensor("op_2665_cast_fp16")]; tensor var_2668 = const()[name = tensor("op_2668"), val = tensor([1, 20, 64, 448])]; tensor var_2669_cast_fp16 = reshape(shape = var_2668, x = key_45_cast_fp16)[name = tensor("op_2669_cast_fp16")]; tensor mh_w_67_transpose_x_0 = const()[name = tensor("mh_w_67_transpose_x_0"), val = tensor(true)]; tensor mh_w_67_transpose_y_0 = const()[name = tensor("mh_w_67_transpose_y_0"), val = tensor(false)]; tensor mh_w_67_cast_fp16 = matmul(transpose_x = mh_w_67_transpose_x_0, transpose_y = mh_w_67_transpose_y_0, x = var_2665_cast_fp16, y = var_2669_cast_fp16)[name = tensor("mh_w_67_cast_fp16")]; tensor mh_w_69_cast_fp16 = add(x = mh_w_67_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_69_cast_fp16")]; tensor var_2677_cast_fp16 = softmax(axis = var_2589, x = mh_w_69_cast_fp16)[name = tensor("op_2677_cast_fp16")]; tensor var_2678 = const()[name = tensor("op_2678"), val = tensor([1, 20, 64, 448])]; tensor var_2679_cast_fp16 = reshape(shape = var_2678, x = value_45_cast_fp16)[name = tensor("op_2679_cast_fp16")]; tensor attn_45_transpose_x_0 = const()[name = tensor("attn_45_transpose_x_0"), val = tensor(false)]; tensor attn_45_transpose_y_0 = const()[name = tensor("attn_45_transpose_y_0"), val = tensor(true)]; tensor attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2679_cast_fp16, y = var_2677_cast_fp16)[name = tensor("attn_45_cast_fp16")]; tensor var_2682 = const()[name = tensor("op_2682"), val = tensor([1, 1280, 1, 1])]; tensor input_111_cast_fp16 = reshape(shape = var_2682, x = attn_45_cast_fp16)[name = tensor("input_111_cast_fp16")]; tensor obj_161_pad_type_0 = const()[name = tensor("obj_161_pad_type_0"), val = tensor("valid")]; tensor obj_161_strides_0 = const()[name = tensor("obj_161_strides_0"), val = tensor([1, 1])]; tensor obj_161_pad_0 = const()[name = tensor("obj_161_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_161_dilations_0 = const()[name = tensor("obj_161_dilations_0"), val = tensor([1, 1])]; tensor obj_161_groups_0 = const()[name = tensor("obj_161_groups_0"), val = tensor(1)]; tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(720980224)))]; tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(724257088)))]; tensor obj_161_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_161_dilations_0, groups = obj_161_groups_0, pad = obj_161_pad_0, pad_type = obj_161_pad_type_0, strides = obj_161_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_111_cast_fp16)[name = tensor("obj_161_cast_fp16")]; tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = obj_161_cast_fp16)[name = tensor("inputs_69_cast_fp16")]; tensor out_69_axes_0 = const()[name = tensor("out_69_axes_0"), val = tensor([1])]; tensor var_2704_to_fp16 = const()[name = tensor("op_2704_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2704_to_fp16, x = inputs_69_cast_fp16)[name = tensor("out_69_cast_fp16")]; tensor obj_163_gamma_0_to_fp16 = const()[name = tensor("obj_163_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(724259712)))]; tensor obj_163_beta_0_to_fp16 = const()[name = tensor("obj_163_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(724262336)))]; tensor obj_163_epsilon_0_to_fp16 = const()[name = tensor("obj_163_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_163_cast_fp16 = batch_norm(beta = obj_163_beta_0_to_fp16, epsilon = obj_163_epsilon_0_to_fp16, gamma = obj_163_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = tensor("obj_163_cast_fp16")]; tensor query_47_pad_type_0 = const()[name = tensor("query_47_pad_type_0"), val = tensor("valid")]; tensor query_47_strides_0 = const()[name = tensor("query_47_strides_0"), val = tensor([1, 1])]; tensor query_47_pad_0 = const()[name = tensor("query_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_47_dilations_0 = const()[name = tensor("query_47_dilations_0"), val = tensor([1, 1])]; tensor query_47_groups_0 = const()[name = tensor("query_47_groups_0"), val = tensor(1)]; tensor layers_11_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_11_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(724264960)))]; tensor layers_11_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_11_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(727541824)))]; tensor query_47_cast_fp16 = conv(bias = layers_11_encoder_attn_q_proj_bias_to_fp16, dilations = query_47_dilations_0, groups = query_47_groups_0, pad = query_47_pad_0, pad_type = query_47_pad_type_0, strides = query_47_strides_0, weight = layers_11_encoder_attn_q_proj_weight_to_fp16, x = obj_163_cast_fp16)[name = tensor("query_47_cast_fp16")]; tensor key_47_pad_type_0 = const()[name = tensor("key_47_pad_type_0"), val = tensor("valid")]; tensor key_47_strides_0 = const()[name = tensor("key_47_strides_0"), val = tensor([1, 1])]; tensor key_47_pad_0 = const()[name = tensor("key_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_47_dilations_0 = const()[name = tensor("key_47_dilations_0"), val = tensor([1, 1])]; tensor key_47_groups_0 = const()[name = tensor("key_47_groups_0"), val = tensor(1)]; tensor layers_11_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_11_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(727544448)))]; tensor key_47_cast_fp16 = conv(dilations = key_47_dilations_0, groups = key_47_groups_0, pad = key_47_pad_0, pad_type = key_47_pad_type_0, strides = key_47_strides_0, weight = layers_11_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_47_cast_fp16")]; tensor value_47_pad_type_0 = const()[name = tensor("value_47_pad_type_0"), val = tensor("valid")]; tensor value_47_strides_0 = const()[name = tensor("value_47_strides_0"), val = tensor([1, 1])]; tensor value_47_pad_0 = const()[name = tensor("value_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_47_dilations_0 = const()[name = tensor("value_47_dilations_0"), val = tensor([1, 1])]; tensor value_47_groups_0 = const()[name = tensor("value_47_groups_0"), val = tensor(1)]; tensor layers_11_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_11_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(730821312)))]; tensor layers_11_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_11_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(734098176)))]; tensor value_47_cast_fp16 = conv(bias = layers_11_encoder_attn_v_proj_bias_to_fp16, dilations = value_47_dilations_0, groups = value_47_groups_0, pad = value_47_pad_0, pad_type = value_47_pad_type_0, strides = value_47_strides_0, weight = layers_11_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_47_cast_fp16")]; tensor var_2740 = const()[name = tensor("op_2740"), val = tensor([1, 20, 64, 1])]; tensor mh_q_47_cast_fp16 = reshape(shape = var_2740, x = query_47_cast_fp16)[name = tensor("mh_q_47_cast_fp16")]; tensor var_2742_to_fp16 = const()[name = tensor("op_2742_to_fp16"), val = tensor(0x1p-3)]; tensor var_2743_cast_fp16 = mul(x = mh_q_47_cast_fp16, y = var_2742_to_fp16)[name = tensor("op_2743_cast_fp16")]; tensor var_2746 = const()[name = tensor("op_2746"), val = tensor([1, 20, 64, 1500])]; tensor var_2747_cast_fp16 = reshape(shape = var_2746, x = key_47_cast_fp16)[name = tensor("op_2747_cast_fp16")]; tensor mh_w_71_transpose_x_0 = const()[name = tensor("mh_w_71_transpose_x_0"), val = tensor(true)]; tensor mh_w_71_transpose_y_0 = const()[name = tensor("mh_w_71_transpose_y_0"), val = tensor(false)]; tensor mh_w_71_cast_fp16 = matmul(transpose_x = mh_w_71_transpose_x_0, transpose_y = mh_w_71_transpose_y_0, x = var_2743_cast_fp16, y = var_2747_cast_fp16)[name = tensor("mh_w_71_cast_fp16")]; tensor obj_167_cast_fp16 = softmax(axis = var_2589, x = mh_w_71_cast_fp16)[name = tensor("obj_167_cast_fp16")]; tensor var_2751 = const()[name = tensor("op_2751"), val = tensor([1, 20, 64, 1500])]; tensor var_2752_cast_fp16 = reshape(shape = var_2751, x = value_47_cast_fp16)[name = tensor("op_2752_cast_fp16")]; tensor attn_47_transpose_x_0 = const()[name = tensor("attn_47_transpose_x_0"), val = tensor(false)]; tensor attn_47_transpose_y_0 = const()[name = tensor("attn_47_transpose_y_0"), val = tensor(true)]; tensor attn_47_cast_fp16 = matmul(transpose_x = attn_47_transpose_x_0, transpose_y = attn_47_transpose_y_0, x = var_2752_cast_fp16, y = obj_167_cast_fp16)[name = tensor("attn_47_cast_fp16")]; tensor var_2755 = const()[name = tensor("op_2755"), val = tensor([1, 1280, 1, 1])]; tensor input_113_cast_fp16 = reshape(shape = var_2755, x = attn_47_cast_fp16)[name = tensor("input_113_cast_fp16")]; tensor obj_165_pad_type_0 = const()[name = tensor("obj_165_pad_type_0"), val = tensor("valid")]; tensor obj_165_strides_0 = const()[name = tensor("obj_165_strides_0"), val = tensor([1, 1])]; tensor obj_165_pad_0 = const()[name = tensor("obj_165_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_165_dilations_0 = const()[name = tensor("obj_165_dilations_0"), val = tensor([1, 1])]; tensor obj_165_groups_0 = const()[name = tensor("obj_165_groups_0"), val = tensor(1)]; tensor layers_11_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_11_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(734100800)))]; tensor layers_11_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_11_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737377664)))]; tensor obj_165_cast_fp16 = conv(bias = layers_11_encoder_attn_o_proj_bias_to_fp16, dilations = obj_165_dilations_0, groups = obj_165_groups_0, pad = obj_165_pad_0, pad_type = obj_165_pad_type_0, strides = obj_165_strides_0, weight = layers_11_encoder_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = tensor("obj_165_cast_fp16")]; tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_165_cast_fp16)[name = tensor("inputs_71_cast_fp16")]; tensor out_71_axes_0 = const()[name = tensor("out_71_axes_0"), val = tensor([1])]; tensor var_2773_to_fp16 = const()[name = tensor("op_2773_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2773_to_fp16, x = inputs_71_cast_fp16)[name = tensor("out_71_cast_fp16")]; tensor input_115_gamma_0_to_fp16 = const()[name = tensor("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737380288)))]; tensor input_115_beta_0_to_fp16 = const()[name = tensor("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737382912)))]; tensor input_115_epsilon_0_to_fp16 = const()[name = tensor("input_115_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = tensor("input_115_cast_fp16")]; tensor input_117_pad_type_0 = const()[name = tensor("input_117_pad_type_0"), val = tensor("valid")]; tensor input_117_strides_0 = const()[name = tensor("input_117_strides_0"), val = tensor([1, 1])]; tensor input_117_pad_0 = const()[name = tensor("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_117_dilations_0 = const()[name = tensor("input_117_dilations_0"), val = tensor([1, 1])]; tensor input_117_groups_0 = const()[name = tensor("input_117_groups_0"), val = tensor(1)]; tensor layers_11_fc1_weight_to_fp16 = const()[name = tensor("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737385536)))]; tensor layers_11_fc1_bias_to_fp16 = const()[name = tensor("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(750492800)))]; tensor input_117_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = tensor("input_117_cast_fp16")]; tensor input_119_mode_0 = const()[name = tensor("input_119_mode_0"), val = tensor("EXACT")]; tensor input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = input_117_cast_fp16)[name = tensor("input_119_cast_fp16")]; tensor hidden_states_25_pad_type_0 = const()[name = tensor("hidden_states_25_pad_type_0"), val = tensor("valid")]; tensor hidden_states_25_strides_0 = const()[name = tensor("hidden_states_25_strides_0"), val = tensor([1, 1])]; tensor hidden_states_25_pad_0 = const()[name = tensor("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_25_dilations_0 = const()[name = tensor("hidden_states_25_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_25_groups_0 = const()[name = tensor("hidden_states_25_groups_0"), val = tensor(1)]; tensor layers_11_fc2_weight_to_fp16 = const()[name = tensor("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(750503104)))]; tensor layers_11_fc2_bias_to_fp16 = const()[name = tensor("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763610368)))]; tensor hidden_states_25_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_119_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_25_cast_fp16)[name = tensor("inputs_73_cast_fp16")]; tensor var_2808 = const()[name = tensor("op_2808"), val = tensor(3)]; tensor out_73_axes_0 = const()[name = tensor("out_73_axes_0"), val = tensor([1])]; tensor var_2833_to_fp16 = const()[name = tensor("op_2833_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_2833_to_fp16, x = inputs_73_cast_fp16)[name = tensor("out_73_cast_fp16")]; tensor obj_169_gamma_0_to_fp16 = const()[name = tensor("obj_169_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763612992)))]; tensor obj_169_beta_0_to_fp16 = const()[name = tensor("obj_169_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763615616)))]; tensor obj_169_epsilon_0_to_fp16 = const()[name = tensor("obj_169_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_169_cast_fp16 = batch_norm(beta = obj_169_beta_0_to_fp16, epsilon = obj_169_epsilon_0_to_fp16, gamma = obj_169_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_73_cast_fp16)[name = tensor("obj_169_cast_fp16")]; tensor query_49_pad_type_0 = const()[name = tensor("query_49_pad_type_0"), val = tensor("valid")]; tensor query_49_strides_0 = const()[name = tensor("query_49_strides_0"), val = tensor([1, 1])]; tensor query_49_pad_0 = const()[name = tensor("query_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_49_dilations_0 = const()[name = tensor("query_49_dilations_0"), val = tensor([1, 1])]; tensor query_49_groups_0 = const()[name = tensor("query_49_groups_0"), val = tensor(1)]; tensor layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763618240)))]; tensor layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(766895104)))]; tensor query_49_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_bias_to_fp16, dilations = query_49_dilations_0, groups = query_49_groups_0, pad = query_49_pad_0, pad_type = query_49_pad_type_0, strides = query_49_strides_0, weight = layers_12_self_attn_q_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = tensor("query_49_cast_fp16")]; tensor current_key_25_pad_type_0 = const()[name = tensor("current_key_25_pad_type_0"), val = tensor("valid")]; tensor current_key_25_strides_0 = const()[name = tensor("current_key_25_strides_0"), val = tensor([1, 1])]; tensor current_key_25_pad_0 = const()[name = tensor("current_key_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_25_dilations_0 = const()[name = tensor("current_key_25_dilations_0"), val = tensor([1, 1])]; tensor current_key_25_groups_0 = const()[name = tensor("current_key_25_groups_0"), val = tensor(1)]; tensor layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(766897728)))]; tensor current_key_25_cast_fp16 = conv(dilations = current_key_25_dilations_0, groups = current_key_25_groups_0, pad = current_key_25_pad_0, pad_type = current_key_25_pad_type_0, strides = current_key_25_strides_0, weight = layers_12_self_attn_k_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = tensor("current_key_25_cast_fp16")]; tensor current_value_25_pad_type_0 = const()[name = tensor("current_value_25_pad_type_0"), val = tensor("valid")]; tensor current_value_25_strides_0 = const()[name = tensor("current_value_25_strides_0"), val = tensor([1, 1])]; tensor current_value_25_pad_0 = const()[name = tensor("current_value_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_25_dilations_0 = const()[name = tensor("current_value_25_dilations_0"), val = tensor([1, 1])]; tensor current_value_25_groups_0 = const()[name = tensor("current_value_25_groups_0"), val = tensor(1)]; tensor layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(770174592)))]; tensor layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773451456)))]; tensor current_value_25_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_bias_to_fp16, dilations = current_value_25_dilations_0, groups = current_value_25_groups_0, pad = current_value_25_pad_0, pad_type = current_value_25_pad_type_0, strides = current_value_25_strides_0, weight = layers_12_self_attn_v_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = tensor("current_value_25_cast_fp16")]; tensor var_2872_cast_fp16 = mul(x = var_103_cast_fp16_12, y = var_239_cast_fp16)[name = tensor("op_2872_cast_fp16")]; tensor var_2873_cast_fp16 = mul(x = current_key_25_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_2873_cast_fp16")]; tensor key_49_cast_fp16 = add(x = var_2872_cast_fp16, y = var_2873_cast_fp16)[name = tensor("key_49_cast_fp16")]; tensor var_2876_cast_fp16 = mul(x = var_138_cast_fp16_12, y = var_239_cast_fp16)[name = tensor("op_2876_cast_fp16")]; tensor var_2877_cast_fp16 = mul(x = current_value_25_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_2877_cast_fp16")]; tensor value_49_cast_fp16 = add(x = var_2876_cast_fp16, y = var_2877_cast_fp16)[name = tensor("value_49_cast_fp16")]; tensor var_2881 = const()[name = tensor("op_2881"), val = tensor([1, 20, 64, 1])]; tensor mh_q_49_cast_fp16 = reshape(shape = var_2881, x = query_49_cast_fp16)[name = tensor("mh_q_49_cast_fp16")]; tensor var_2883_to_fp16 = const()[name = tensor("op_2883_to_fp16"), val = tensor(0x1p-3)]; tensor var_2884_cast_fp16 = mul(x = mh_q_49_cast_fp16, y = var_2883_to_fp16)[name = tensor("op_2884_cast_fp16")]; tensor var_2887 = const()[name = tensor("op_2887"), val = tensor([1, 20, 64, 448])]; tensor var_2888_cast_fp16 = reshape(shape = var_2887, x = key_49_cast_fp16)[name = tensor("op_2888_cast_fp16")]; tensor mh_w_73_transpose_x_0 = const()[name = tensor("mh_w_73_transpose_x_0"), val = tensor(true)]; tensor mh_w_73_transpose_y_0 = const()[name = tensor("mh_w_73_transpose_y_0"), val = tensor(false)]; tensor mh_w_73_cast_fp16 = matmul(transpose_x = mh_w_73_transpose_x_0, transpose_y = mh_w_73_transpose_y_0, x = var_2884_cast_fp16, y = var_2888_cast_fp16)[name = tensor("mh_w_73_cast_fp16")]; tensor mh_w_75_cast_fp16 = add(x = mh_w_73_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_75_cast_fp16")]; tensor var_2896_cast_fp16 = softmax(axis = var_2808, x = mh_w_75_cast_fp16)[name = tensor("op_2896_cast_fp16")]; tensor var_2897 = const()[name = tensor("op_2897"), val = tensor([1, 20, 64, 448])]; tensor var_2898_cast_fp16 = reshape(shape = var_2897, x = value_49_cast_fp16)[name = tensor("op_2898_cast_fp16")]; tensor attn_49_transpose_x_0 = const()[name = tensor("attn_49_transpose_x_0"), val = tensor(false)]; tensor attn_49_transpose_y_0 = const()[name = tensor("attn_49_transpose_y_0"), val = tensor(true)]; tensor attn_49_cast_fp16 = matmul(transpose_x = attn_49_transpose_x_0, transpose_y = attn_49_transpose_y_0, x = var_2898_cast_fp16, y = var_2896_cast_fp16)[name = tensor("attn_49_cast_fp16")]; tensor var_2901 = const()[name = tensor("op_2901"), val = tensor([1, 1280, 1, 1])]; tensor input_121_cast_fp16 = reshape(shape = var_2901, x = attn_49_cast_fp16)[name = tensor("input_121_cast_fp16")]; tensor obj_175_pad_type_0 = const()[name = tensor("obj_175_pad_type_0"), val = tensor("valid")]; tensor obj_175_strides_0 = const()[name = tensor("obj_175_strides_0"), val = tensor([1, 1])]; tensor obj_175_pad_0 = const()[name = tensor("obj_175_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_175_dilations_0 = const()[name = tensor("obj_175_dilations_0"), val = tensor([1, 1])]; tensor obj_175_groups_0 = const()[name = tensor("obj_175_groups_0"), val = tensor(1)]; tensor layers_12_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_12_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773454080)))]; tensor layers_12_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(776730944)))]; tensor obj_175_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_bias_to_fp16, dilations = obj_175_dilations_0, groups = obj_175_groups_0, pad = obj_175_pad_0, pad_type = obj_175_pad_type_0, strides = obj_175_strides_0, weight = layers_12_self_attn_o_proj_weight_to_fp16, x = input_121_cast_fp16)[name = tensor("obj_175_cast_fp16")]; tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_175_cast_fp16)[name = tensor("inputs_75_cast_fp16")]; tensor out_75_axes_0 = const()[name = tensor("out_75_axes_0"), val = tensor([1])]; tensor var_2923_to_fp16 = const()[name = tensor("op_2923_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_2923_to_fp16, x = inputs_75_cast_fp16)[name = tensor("out_75_cast_fp16")]; tensor obj_177_gamma_0_to_fp16 = const()[name = tensor("obj_177_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(776733568)))]; tensor obj_177_beta_0_to_fp16 = const()[name = tensor("obj_177_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(776736192)))]; tensor obj_177_epsilon_0_to_fp16 = const()[name = tensor("obj_177_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_177_cast_fp16 = batch_norm(beta = obj_177_beta_0_to_fp16, epsilon = obj_177_epsilon_0_to_fp16, gamma = obj_177_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_75_cast_fp16)[name = tensor("obj_177_cast_fp16")]; tensor query_51_pad_type_0 = const()[name = tensor("query_51_pad_type_0"), val = tensor("valid")]; tensor query_51_strides_0 = const()[name = tensor("query_51_strides_0"), val = tensor([1, 1])]; tensor query_51_pad_0 = const()[name = tensor("query_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_51_dilations_0 = const()[name = tensor("query_51_dilations_0"), val = tensor([1, 1])]; tensor query_51_groups_0 = const()[name = tensor("query_51_groups_0"), val = tensor(1)]; tensor layers_12_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_12_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(776738816)))]; tensor layers_12_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_12_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(780015680)))]; tensor query_51_cast_fp16 = conv(bias = layers_12_encoder_attn_q_proj_bias_to_fp16, dilations = query_51_dilations_0, groups = query_51_groups_0, pad = query_51_pad_0, pad_type = query_51_pad_type_0, strides = query_51_strides_0, weight = layers_12_encoder_attn_q_proj_weight_to_fp16, x = obj_177_cast_fp16)[name = tensor("query_51_cast_fp16")]; tensor key_51_pad_type_0 = const()[name = tensor("key_51_pad_type_0"), val = tensor("valid")]; tensor key_51_strides_0 = const()[name = tensor("key_51_strides_0"), val = tensor([1, 1])]; tensor key_51_pad_0 = const()[name = tensor("key_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_51_dilations_0 = const()[name = tensor("key_51_dilations_0"), val = tensor([1, 1])]; tensor key_51_groups_0 = const()[name = tensor("key_51_groups_0"), val = tensor(1)]; tensor layers_12_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_12_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(780018304)))]; tensor key_51_cast_fp16 = conv(dilations = key_51_dilations_0, groups = key_51_groups_0, pad = key_51_pad_0, pad_type = key_51_pad_type_0, strides = key_51_strides_0, weight = layers_12_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_51_cast_fp16")]; tensor value_51_pad_type_0 = const()[name = tensor("value_51_pad_type_0"), val = tensor("valid")]; tensor value_51_strides_0 = const()[name = tensor("value_51_strides_0"), val = tensor([1, 1])]; tensor value_51_pad_0 = const()[name = tensor("value_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_51_dilations_0 = const()[name = tensor("value_51_dilations_0"), val = tensor([1, 1])]; tensor value_51_groups_0 = const()[name = tensor("value_51_groups_0"), val = tensor(1)]; tensor layers_12_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_12_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(783295168)))]; tensor layers_12_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_12_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(786572032)))]; tensor value_51_cast_fp16 = conv(bias = layers_12_encoder_attn_v_proj_bias_to_fp16, dilations = value_51_dilations_0, groups = value_51_groups_0, pad = value_51_pad_0, pad_type = value_51_pad_type_0, strides = value_51_strides_0, weight = layers_12_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_51_cast_fp16")]; tensor var_2959 = const()[name = tensor("op_2959"), val = tensor([1, 20, 64, 1])]; tensor mh_q_51_cast_fp16 = reshape(shape = var_2959, x = query_51_cast_fp16)[name = tensor("mh_q_51_cast_fp16")]; tensor var_2961_to_fp16 = const()[name = tensor("op_2961_to_fp16"), val = tensor(0x1p-3)]; tensor var_2962_cast_fp16 = mul(x = mh_q_51_cast_fp16, y = var_2961_to_fp16)[name = tensor("op_2962_cast_fp16")]; tensor var_2965 = const()[name = tensor("op_2965"), val = tensor([1, 20, 64, 1500])]; tensor var_2966_cast_fp16 = reshape(shape = var_2965, x = key_51_cast_fp16)[name = tensor("op_2966_cast_fp16")]; tensor mh_w_77_transpose_x_0 = const()[name = tensor("mh_w_77_transpose_x_0"), val = tensor(true)]; tensor mh_w_77_transpose_y_0 = const()[name = tensor("mh_w_77_transpose_y_0"), val = tensor(false)]; tensor mh_w_77_cast_fp16 = matmul(transpose_x = mh_w_77_transpose_x_0, transpose_y = mh_w_77_transpose_y_0, x = var_2962_cast_fp16, y = var_2966_cast_fp16)[name = tensor("mh_w_77_cast_fp16")]; tensor obj_181_cast_fp16 = softmax(axis = var_2808, x = mh_w_77_cast_fp16)[name = tensor("obj_181_cast_fp16")]; tensor var_2970 = const()[name = tensor("op_2970"), val = tensor([1, 20, 64, 1500])]; tensor var_2971_cast_fp16 = reshape(shape = var_2970, x = value_51_cast_fp16)[name = tensor("op_2971_cast_fp16")]; tensor attn_51_transpose_x_0 = const()[name = tensor("attn_51_transpose_x_0"), val = tensor(false)]; tensor attn_51_transpose_y_0 = const()[name = tensor("attn_51_transpose_y_0"), val = tensor(true)]; tensor attn_51_cast_fp16 = matmul(transpose_x = attn_51_transpose_x_0, transpose_y = attn_51_transpose_y_0, x = var_2971_cast_fp16, y = obj_181_cast_fp16)[name = tensor("attn_51_cast_fp16")]; tensor var_2974 = const()[name = tensor("op_2974"), val = tensor([1, 1280, 1, 1])]; tensor input_123_cast_fp16 = reshape(shape = var_2974, x = attn_51_cast_fp16)[name = tensor("input_123_cast_fp16")]; tensor obj_179_pad_type_0 = const()[name = tensor("obj_179_pad_type_0"), val = tensor("valid")]; tensor obj_179_strides_0 = const()[name = tensor("obj_179_strides_0"), val = tensor([1, 1])]; tensor obj_179_pad_0 = const()[name = tensor("obj_179_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_179_dilations_0 = const()[name = tensor("obj_179_dilations_0"), val = tensor([1, 1])]; tensor obj_179_groups_0 = const()[name = tensor("obj_179_groups_0"), val = tensor(1)]; tensor layers_12_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_12_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(786574656)))]; tensor layers_12_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_12_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789851520)))]; tensor obj_179_cast_fp16 = conv(bias = layers_12_encoder_attn_o_proj_bias_to_fp16, dilations = obj_179_dilations_0, groups = obj_179_groups_0, pad = obj_179_pad_0, pad_type = obj_179_pad_type_0, strides = obj_179_strides_0, weight = layers_12_encoder_attn_o_proj_weight_to_fp16, x = input_123_cast_fp16)[name = tensor("obj_179_cast_fp16")]; tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = obj_179_cast_fp16)[name = tensor("inputs_77_cast_fp16")]; tensor out_77_axes_0 = const()[name = tensor("out_77_axes_0"), val = tensor([1])]; tensor var_2992_to_fp16 = const()[name = tensor("op_2992_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_2992_to_fp16, x = inputs_77_cast_fp16)[name = tensor("out_77_cast_fp16")]; tensor input_125_gamma_0_to_fp16 = const()[name = tensor("input_125_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789854144)))]; tensor input_125_beta_0_to_fp16 = const()[name = tensor("input_125_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789856768)))]; tensor input_125_epsilon_0_to_fp16 = const()[name = tensor("input_125_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_125_cast_fp16 = batch_norm(beta = input_125_beta_0_to_fp16, epsilon = input_125_epsilon_0_to_fp16, gamma = input_125_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_77_cast_fp16)[name = tensor("input_125_cast_fp16")]; tensor input_127_pad_type_0 = const()[name = tensor("input_127_pad_type_0"), val = tensor("valid")]; tensor input_127_strides_0 = const()[name = tensor("input_127_strides_0"), val = tensor([1, 1])]; tensor input_127_pad_0 = const()[name = tensor("input_127_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_127_dilations_0 = const()[name = tensor("input_127_dilations_0"), val = tensor([1, 1])]; tensor input_127_groups_0 = const()[name = tensor("input_127_groups_0"), val = tensor(1)]; tensor layers_12_fc1_weight_to_fp16 = const()[name = tensor("layers_12_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(789859392)))]; tensor layers_12_fc1_bias_to_fp16 = const()[name = tensor("layers_12_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802966656)))]; tensor input_127_cast_fp16 = conv(bias = layers_12_fc1_bias_to_fp16, dilations = input_127_dilations_0, groups = input_127_groups_0, pad = input_127_pad_0, pad_type = input_127_pad_type_0, strides = input_127_strides_0, weight = layers_12_fc1_weight_to_fp16, x = input_125_cast_fp16)[name = tensor("input_127_cast_fp16")]; tensor input_129_mode_0 = const()[name = tensor("input_129_mode_0"), val = tensor("EXACT")]; tensor input_129_cast_fp16 = gelu(mode = input_129_mode_0, x = input_127_cast_fp16)[name = tensor("input_129_cast_fp16")]; tensor hidden_states_27_pad_type_0 = const()[name = tensor("hidden_states_27_pad_type_0"), val = tensor("valid")]; tensor hidden_states_27_strides_0 = const()[name = tensor("hidden_states_27_strides_0"), val = tensor([1, 1])]; tensor hidden_states_27_pad_0 = const()[name = tensor("hidden_states_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_27_dilations_0 = const()[name = tensor("hidden_states_27_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_27_groups_0 = const()[name = tensor("hidden_states_27_groups_0"), val = tensor(1)]; tensor layers_12_fc2_weight_to_fp16 = const()[name = tensor("layers_12_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(802976960)))]; tensor layers_12_fc2_bias_to_fp16 = const()[name = tensor("layers_12_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(816084224)))]; tensor hidden_states_27_cast_fp16 = conv(bias = layers_12_fc2_bias_to_fp16, dilations = hidden_states_27_dilations_0, groups = hidden_states_27_groups_0, pad = hidden_states_27_pad_0, pad_type = hidden_states_27_pad_type_0, strides = hidden_states_27_strides_0, weight = layers_12_fc2_weight_to_fp16, x = input_129_cast_fp16)[name = tensor("hidden_states_27_cast_fp16")]; tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = hidden_states_27_cast_fp16)[name = tensor("inputs_79_cast_fp16")]; tensor var_3027 = const()[name = tensor("op_3027"), val = tensor(3)]; tensor out_79_axes_0 = const()[name = tensor("out_79_axes_0"), val = tensor([1])]; tensor var_3052_to_fp16 = const()[name = tensor("op_3052_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_3052_to_fp16, x = inputs_79_cast_fp16)[name = tensor("out_79_cast_fp16")]; tensor obj_183_gamma_0_to_fp16 = const()[name = tensor("obj_183_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(816086848)))]; tensor obj_183_beta_0_to_fp16 = const()[name = tensor("obj_183_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(816089472)))]; tensor obj_183_epsilon_0_to_fp16 = const()[name = tensor("obj_183_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_183_cast_fp16 = batch_norm(beta = obj_183_beta_0_to_fp16, epsilon = obj_183_epsilon_0_to_fp16, gamma = obj_183_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_79_cast_fp16)[name = tensor("obj_183_cast_fp16")]; tensor query_53_pad_type_0 = const()[name = tensor("query_53_pad_type_0"), val = tensor("valid")]; tensor query_53_strides_0 = const()[name = tensor("query_53_strides_0"), val = tensor([1, 1])]; tensor query_53_pad_0 = const()[name = tensor("query_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_53_dilations_0 = const()[name = tensor("query_53_dilations_0"), val = tensor([1, 1])]; tensor query_53_groups_0 = const()[name = tensor("query_53_groups_0"), val = tensor(1)]; tensor layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(816092096)))]; tensor layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819368960)))]; tensor query_53_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_bias_to_fp16, dilations = query_53_dilations_0, groups = query_53_groups_0, pad = query_53_pad_0, pad_type = query_53_pad_type_0, strides = query_53_strides_0, weight = layers_13_self_attn_q_proj_weight_to_fp16, x = obj_183_cast_fp16)[name = tensor("query_53_cast_fp16")]; tensor current_key_27_pad_type_0 = const()[name = tensor("current_key_27_pad_type_0"), val = tensor("valid")]; tensor current_key_27_strides_0 = const()[name = tensor("current_key_27_strides_0"), val = tensor([1, 1])]; tensor current_key_27_pad_0 = const()[name = tensor("current_key_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_27_dilations_0 = const()[name = tensor("current_key_27_dilations_0"), val = tensor([1, 1])]; tensor current_key_27_groups_0 = const()[name = tensor("current_key_27_groups_0"), val = tensor(1)]; tensor layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(819371584)))]; tensor current_key_27_cast_fp16 = conv(dilations = current_key_27_dilations_0, groups = current_key_27_groups_0, pad = current_key_27_pad_0, pad_type = current_key_27_pad_type_0, strides = current_key_27_strides_0, weight = layers_13_self_attn_k_proj_weight_to_fp16, x = obj_183_cast_fp16)[name = tensor("current_key_27_cast_fp16")]; tensor current_value_27_pad_type_0 = const()[name = tensor("current_value_27_pad_type_0"), val = tensor("valid")]; tensor current_value_27_strides_0 = const()[name = tensor("current_value_27_strides_0"), val = tensor([1, 1])]; tensor current_value_27_pad_0 = const()[name = tensor("current_value_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_27_dilations_0 = const()[name = tensor("current_value_27_dilations_0"), val = tensor([1, 1])]; tensor current_value_27_groups_0 = const()[name = tensor("current_value_27_groups_0"), val = tensor(1)]; tensor layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(822648448)))]; tensor layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(825925312)))]; tensor current_value_27_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_bias_to_fp16, dilations = current_value_27_dilations_0, groups = current_value_27_groups_0, pad = current_value_27_pad_0, pad_type = current_value_27_pad_type_0, strides = current_value_27_strides_0, weight = layers_13_self_attn_v_proj_weight_to_fp16, x = obj_183_cast_fp16)[name = tensor("current_value_27_cast_fp16")]; tensor var_3091_cast_fp16 = mul(x = var_103_cast_fp16_13, y = var_239_cast_fp16)[name = tensor("op_3091_cast_fp16")]; tensor var_3092_cast_fp16 = mul(x = current_key_27_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3092_cast_fp16")]; tensor key_53_cast_fp16 = add(x = var_3091_cast_fp16, y = var_3092_cast_fp16)[name = tensor("key_53_cast_fp16")]; tensor var_3095_cast_fp16 = mul(x = var_138_cast_fp16_13, y = var_239_cast_fp16)[name = tensor("op_3095_cast_fp16")]; tensor var_3096_cast_fp16 = mul(x = current_value_27_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3096_cast_fp16")]; tensor value_53_cast_fp16 = add(x = var_3095_cast_fp16, y = var_3096_cast_fp16)[name = tensor("value_53_cast_fp16")]; tensor var_3100 = const()[name = tensor("op_3100"), val = tensor([1, 20, 64, 1])]; tensor mh_q_53_cast_fp16 = reshape(shape = var_3100, x = query_53_cast_fp16)[name = tensor("mh_q_53_cast_fp16")]; tensor var_3102_to_fp16 = const()[name = tensor("op_3102_to_fp16"), val = tensor(0x1p-3)]; tensor var_3103_cast_fp16 = mul(x = mh_q_53_cast_fp16, y = var_3102_to_fp16)[name = tensor("op_3103_cast_fp16")]; tensor var_3106 = const()[name = tensor("op_3106"), val = tensor([1, 20, 64, 448])]; tensor var_3107_cast_fp16 = reshape(shape = var_3106, x = key_53_cast_fp16)[name = tensor("op_3107_cast_fp16")]; tensor mh_w_79_transpose_x_0 = const()[name = tensor("mh_w_79_transpose_x_0"), val = tensor(true)]; tensor mh_w_79_transpose_y_0 = const()[name = tensor("mh_w_79_transpose_y_0"), val = tensor(false)]; tensor mh_w_79_cast_fp16 = matmul(transpose_x = mh_w_79_transpose_x_0, transpose_y = mh_w_79_transpose_y_0, x = var_3103_cast_fp16, y = var_3107_cast_fp16)[name = tensor("mh_w_79_cast_fp16")]; tensor mh_w_81_cast_fp16 = add(x = mh_w_79_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_81_cast_fp16")]; tensor var_3115_cast_fp16 = softmax(axis = var_3027, x = mh_w_81_cast_fp16)[name = tensor("op_3115_cast_fp16")]; tensor var_3116 = const()[name = tensor("op_3116"), val = tensor([1, 20, 64, 448])]; tensor var_3117_cast_fp16 = reshape(shape = var_3116, x = value_53_cast_fp16)[name = tensor("op_3117_cast_fp16")]; tensor attn_53_transpose_x_0 = const()[name = tensor("attn_53_transpose_x_0"), val = tensor(false)]; tensor attn_53_transpose_y_0 = const()[name = tensor("attn_53_transpose_y_0"), val = tensor(true)]; tensor attn_53_cast_fp16 = matmul(transpose_x = attn_53_transpose_x_0, transpose_y = attn_53_transpose_y_0, x = var_3117_cast_fp16, y = var_3115_cast_fp16)[name = tensor("attn_53_cast_fp16")]; tensor var_3120 = const()[name = tensor("op_3120"), val = tensor([1, 1280, 1, 1])]; tensor input_131_cast_fp16 = reshape(shape = var_3120, x = attn_53_cast_fp16)[name = tensor("input_131_cast_fp16")]; tensor obj_189_pad_type_0 = const()[name = tensor("obj_189_pad_type_0"), val = tensor("valid")]; tensor obj_189_strides_0 = const()[name = tensor("obj_189_strides_0"), val = tensor([1, 1])]; tensor obj_189_pad_0 = const()[name = tensor("obj_189_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_189_dilations_0 = const()[name = tensor("obj_189_dilations_0"), val = tensor([1, 1])]; tensor obj_189_groups_0 = const()[name = tensor("obj_189_groups_0"), val = tensor(1)]; tensor layers_13_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_13_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(825927936)))]; tensor layers_13_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(829204800)))]; tensor obj_189_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_bias_to_fp16, dilations = obj_189_dilations_0, groups = obj_189_groups_0, pad = obj_189_pad_0, pad_type = obj_189_pad_type_0, strides = obj_189_strides_0, weight = layers_13_self_attn_o_proj_weight_to_fp16, x = input_131_cast_fp16)[name = tensor("obj_189_cast_fp16")]; tensor inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = obj_189_cast_fp16)[name = tensor("inputs_81_cast_fp16")]; tensor out_81_axes_0 = const()[name = tensor("out_81_axes_0"), val = tensor([1])]; tensor var_3142_to_fp16 = const()[name = tensor("op_3142_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_3142_to_fp16, x = inputs_81_cast_fp16)[name = tensor("out_81_cast_fp16")]; tensor obj_191_gamma_0_to_fp16 = const()[name = tensor("obj_191_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(829207424)))]; tensor obj_191_beta_0_to_fp16 = const()[name = tensor("obj_191_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(829210048)))]; tensor obj_191_epsilon_0_to_fp16 = const()[name = tensor("obj_191_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_191_cast_fp16 = batch_norm(beta = obj_191_beta_0_to_fp16, epsilon = obj_191_epsilon_0_to_fp16, gamma = obj_191_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_81_cast_fp16)[name = tensor("obj_191_cast_fp16")]; tensor query_55_pad_type_0 = const()[name = tensor("query_55_pad_type_0"), val = tensor("valid")]; tensor query_55_strides_0 = const()[name = tensor("query_55_strides_0"), val = tensor([1, 1])]; tensor query_55_pad_0 = const()[name = tensor("query_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_55_dilations_0 = const()[name = tensor("query_55_dilations_0"), val = tensor([1, 1])]; tensor query_55_groups_0 = const()[name = tensor("query_55_groups_0"), val = tensor(1)]; tensor layers_13_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_13_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(829212672)))]; tensor layers_13_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_13_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(832489536)))]; tensor query_55_cast_fp16 = conv(bias = layers_13_encoder_attn_q_proj_bias_to_fp16, dilations = query_55_dilations_0, groups = query_55_groups_0, pad = query_55_pad_0, pad_type = query_55_pad_type_0, strides = query_55_strides_0, weight = layers_13_encoder_attn_q_proj_weight_to_fp16, x = obj_191_cast_fp16)[name = tensor("query_55_cast_fp16")]; tensor key_55_pad_type_0 = const()[name = tensor("key_55_pad_type_0"), val = tensor("valid")]; tensor key_55_strides_0 = const()[name = tensor("key_55_strides_0"), val = tensor([1, 1])]; tensor key_55_pad_0 = const()[name = tensor("key_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_55_dilations_0 = const()[name = tensor("key_55_dilations_0"), val = tensor([1, 1])]; tensor key_55_groups_0 = const()[name = tensor("key_55_groups_0"), val = tensor(1)]; tensor layers_13_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_13_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(832492160)))]; tensor key_55_cast_fp16 = conv(dilations = key_55_dilations_0, groups = key_55_groups_0, pad = key_55_pad_0, pad_type = key_55_pad_type_0, strides = key_55_strides_0, weight = layers_13_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_55_cast_fp16")]; tensor value_55_pad_type_0 = const()[name = tensor("value_55_pad_type_0"), val = tensor("valid")]; tensor value_55_strides_0 = const()[name = tensor("value_55_strides_0"), val = tensor([1, 1])]; tensor value_55_pad_0 = const()[name = tensor("value_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_55_dilations_0 = const()[name = tensor("value_55_dilations_0"), val = tensor([1, 1])]; tensor value_55_groups_0 = const()[name = tensor("value_55_groups_0"), val = tensor(1)]; tensor layers_13_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_13_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835769024)))]; tensor layers_13_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_13_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(839045888)))]; tensor value_55_cast_fp16 = conv(bias = layers_13_encoder_attn_v_proj_bias_to_fp16, dilations = value_55_dilations_0, groups = value_55_groups_0, pad = value_55_pad_0, pad_type = value_55_pad_type_0, strides = value_55_strides_0, weight = layers_13_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_55_cast_fp16")]; tensor var_3178 = const()[name = tensor("op_3178"), val = tensor([1, 20, 64, 1])]; tensor mh_q_55_cast_fp16 = reshape(shape = var_3178, x = query_55_cast_fp16)[name = tensor("mh_q_55_cast_fp16")]; tensor var_3180_to_fp16 = const()[name = tensor("op_3180_to_fp16"), val = tensor(0x1p-3)]; tensor var_3181_cast_fp16 = mul(x = mh_q_55_cast_fp16, y = var_3180_to_fp16)[name = tensor("op_3181_cast_fp16")]; tensor var_3184 = const()[name = tensor("op_3184"), val = tensor([1, 20, 64, 1500])]; tensor var_3185_cast_fp16 = reshape(shape = var_3184, x = key_55_cast_fp16)[name = tensor("op_3185_cast_fp16")]; tensor mh_w_83_transpose_x_0 = const()[name = tensor("mh_w_83_transpose_x_0"), val = tensor(true)]; tensor mh_w_83_transpose_y_0 = const()[name = tensor("mh_w_83_transpose_y_0"), val = tensor(false)]; tensor mh_w_83_cast_fp16 = matmul(transpose_x = mh_w_83_transpose_x_0, transpose_y = mh_w_83_transpose_y_0, x = var_3181_cast_fp16, y = var_3185_cast_fp16)[name = tensor("mh_w_83_cast_fp16")]; tensor obj_195_cast_fp16 = softmax(axis = var_3027, x = mh_w_83_cast_fp16)[name = tensor("obj_195_cast_fp16")]; tensor var_3189 = const()[name = tensor("op_3189"), val = tensor([1, 20, 64, 1500])]; tensor var_3190_cast_fp16 = reshape(shape = var_3189, x = value_55_cast_fp16)[name = tensor("op_3190_cast_fp16")]; tensor attn_55_transpose_x_0 = const()[name = tensor("attn_55_transpose_x_0"), val = tensor(false)]; tensor attn_55_transpose_y_0 = const()[name = tensor("attn_55_transpose_y_0"), val = tensor(true)]; tensor attn_55_cast_fp16 = matmul(transpose_x = attn_55_transpose_x_0, transpose_y = attn_55_transpose_y_0, x = var_3190_cast_fp16, y = obj_195_cast_fp16)[name = tensor("attn_55_cast_fp16")]; tensor var_3193 = const()[name = tensor("op_3193"), val = tensor([1, 1280, 1, 1])]; tensor input_133_cast_fp16 = reshape(shape = var_3193, x = attn_55_cast_fp16)[name = tensor("input_133_cast_fp16")]; tensor obj_193_pad_type_0 = const()[name = tensor("obj_193_pad_type_0"), val = tensor("valid")]; tensor obj_193_strides_0 = const()[name = tensor("obj_193_strides_0"), val = tensor([1, 1])]; tensor obj_193_pad_0 = const()[name = tensor("obj_193_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_193_dilations_0 = const()[name = tensor("obj_193_dilations_0"), val = tensor([1, 1])]; tensor obj_193_groups_0 = const()[name = tensor("obj_193_groups_0"), val = tensor(1)]; tensor layers_13_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_13_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(839048512)))]; tensor layers_13_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_13_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(842325376)))]; tensor obj_193_cast_fp16 = conv(bias = layers_13_encoder_attn_o_proj_bias_to_fp16, dilations = obj_193_dilations_0, groups = obj_193_groups_0, pad = obj_193_pad_0, pad_type = obj_193_pad_type_0, strides = obj_193_strides_0, weight = layers_13_encoder_attn_o_proj_weight_to_fp16, x = input_133_cast_fp16)[name = tensor("obj_193_cast_fp16")]; tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = obj_193_cast_fp16)[name = tensor("inputs_83_cast_fp16")]; tensor out_83_axes_0 = const()[name = tensor("out_83_axes_0"), val = tensor([1])]; tensor var_3214_to_fp16 = const()[name = tensor("op_3214_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_3214_to_fp16, x = inputs_83_cast_fp16)[name = tensor("out_83_cast_fp16")]; tensor input_135_gamma_0_to_fp16 = const()[name = tensor("input_135_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(842328000)))]; tensor input_135_beta_0_to_fp16 = const()[name = tensor("input_135_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(842330624)))]; tensor input_135_epsilon_0_to_fp16 = const()[name = tensor("input_135_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_135_cast_fp16 = batch_norm(beta = input_135_beta_0_to_fp16, epsilon = input_135_epsilon_0_to_fp16, gamma = input_135_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_83_cast_fp16)[name = tensor("input_135_cast_fp16")]; tensor input_137_pad_type_0 = const()[name = tensor("input_137_pad_type_0"), val = tensor("valid")]; tensor input_137_strides_0 = const()[name = tensor("input_137_strides_0"), val = tensor([1, 1])]; tensor input_137_pad_0 = const()[name = tensor("input_137_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_137_dilations_0 = const()[name = tensor("input_137_dilations_0"), val = tensor([1, 1])]; tensor input_137_groups_0 = const()[name = tensor("input_137_groups_0"), val = tensor(1)]; tensor layers_13_fc1_weight_to_fp16 = const()[name = tensor("layers_13_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(842333248)))]; tensor layers_13_fc1_bias_to_fp16 = const()[name = tensor("layers_13_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(855440512)))]; tensor input_137_cast_fp16 = conv(bias = layers_13_fc1_bias_to_fp16, dilations = input_137_dilations_0, groups = input_137_groups_0, pad = input_137_pad_0, pad_type = input_137_pad_type_0, strides = input_137_strides_0, weight = layers_13_fc1_weight_to_fp16, x = input_135_cast_fp16)[name = tensor("input_137_cast_fp16")]; tensor input_139_mode_0 = const()[name = tensor("input_139_mode_0"), val = tensor("EXACT")]; tensor input_139_cast_fp16 = gelu(mode = input_139_mode_0, x = input_137_cast_fp16)[name = tensor("input_139_cast_fp16")]; tensor hidden_states_29_pad_type_0 = const()[name = tensor("hidden_states_29_pad_type_0"), val = tensor("valid")]; tensor hidden_states_29_strides_0 = const()[name = tensor("hidden_states_29_strides_0"), val = tensor([1, 1])]; tensor hidden_states_29_pad_0 = const()[name = tensor("hidden_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_29_dilations_0 = const()[name = tensor("hidden_states_29_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_29_groups_0 = const()[name = tensor("hidden_states_29_groups_0"), val = tensor(1)]; tensor layers_13_fc2_weight_to_fp16 = const()[name = tensor("layers_13_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(855450816)))]; tensor layers_13_fc2_bias_to_fp16 = const()[name = tensor("layers_13_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(868558080)))]; tensor hidden_states_29_cast_fp16 = conv(bias = layers_13_fc2_bias_to_fp16, dilations = hidden_states_29_dilations_0, groups = hidden_states_29_groups_0, pad = hidden_states_29_pad_0, pad_type = hidden_states_29_pad_type_0, strides = hidden_states_29_strides_0, weight = layers_13_fc2_weight_to_fp16, x = input_139_cast_fp16)[name = tensor("hidden_states_29_cast_fp16")]; tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = hidden_states_29_cast_fp16)[name = tensor("inputs_85_cast_fp16")]; tensor var_3250 = const()[name = tensor("op_3250"), val = tensor(3)]; tensor out_85_axes_0 = const()[name = tensor("out_85_axes_0"), val = tensor([1])]; tensor var_3275_to_fp16 = const()[name = tensor("op_3275_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_3275_to_fp16, x = inputs_85_cast_fp16)[name = tensor("out_85_cast_fp16")]; tensor obj_197_gamma_0_to_fp16 = const()[name = tensor("obj_197_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(868560704)))]; tensor obj_197_beta_0_to_fp16 = const()[name = tensor("obj_197_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(868563328)))]; tensor obj_197_epsilon_0_to_fp16 = const()[name = tensor("obj_197_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_197_cast_fp16 = batch_norm(beta = obj_197_beta_0_to_fp16, epsilon = obj_197_epsilon_0_to_fp16, gamma = obj_197_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_85_cast_fp16)[name = tensor("obj_197_cast_fp16")]; tensor query_57_pad_type_0 = const()[name = tensor("query_57_pad_type_0"), val = tensor("valid")]; tensor query_57_strides_0 = const()[name = tensor("query_57_strides_0"), val = tensor([1, 1])]; tensor query_57_pad_0 = const()[name = tensor("query_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_57_dilations_0 = const()[name = tensor("query_57_dilations_0"), val = tensor([1, 1])]; tensor query_57_groups_0 = const()[name = tensor("query_57_groups_0"), val = tensor(1)]; tensor layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(868565952)))]; tensor layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(871842816)))]; tensor query_57_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_bias_to_fp16, dilations = query_57_dilations_0, groups = query_57_groups_0, pad = query_57_pad_0, pad_type = query_57_pad_type_0, strides = query_57_strides_0, weight = layers_14_self_attn_q_proj_weight_to_fp16, x = obj_197_cast_fp16)[name = tensor("query_57_cast_fp16")]; tensor current_key_29_pad_type_0 = const()[name = tensor("current_key_29_pad_type_0"), val = tensor("valid")]; tensor current_key_29_strides_0 = const()[name = tensor("current_key_29_strides_0"), val = tensor([1, 1])]; tensor current_key_29_pad_0 = const()[name = tensor("current_key_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_29_dilations_0 = const()[name = tensor("current_key_29_dilations_0"), val = tensor([1, 1])]; tensor current_key_29_groups_0 = const()[name = tensor("current_key_29_groups_0"), val = tensor(1)]; tensor layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(871845440)))]; tensor current_key_29_cast_fp16 = conv(dilations = current_key_29_dilations_0, groups = current_key_29_groups_0, pad = current_key_29_pad_0, pad_type = current_key_29_pad_type_0, strides = current_key_29_strides_0, weight = layers_14_self_attn_k_proj_weight_to_fp16, x = obj_197_cast_fp16)[name = tensor("current_key_29_cast_fp16")]; tensor current_value_29_pad_type_0 = const()[name = tensor("current_value_29_pad_type_0"), val = tensor("valid")]; tensor current_value_29_strides_0 = const()[name = tensor("current_value_29_strides_0"), val = tensor([1, 1])]; tensor current_value_29_pad_0 = const()[name = tensor("current_value_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_29_dilations_0 = const()[name = tensor("current_value_29_dilations_0"), val = tensor([1, 1])]; tensor current_value_29_groups_0 = const()[name = tensor("current_value_29_groups_0"), val = tensor(1)]; tensor layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(875122304)))]; tensor layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(878399168)))]; tensor current_value_29_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_bias_to_fp16, dilations = current_value_29_dilations_0, groups = current_value_29_groups_0, pad = current_value_29_pad_0, pad_type = current_value_29_pad_type_0, strides = current_value_29_strides_0, weight = layers_14_self_attn_v_proj_weight_to_fp16, x = obj_197_cast_fp16)[name = tensor("current_value_29_cast_fp16")]; tensor var_3314_cast_fp16 = mul(x = var_103_cast_fp16_14, y = var_239_cast_fp16)[name = tensor("op_3314_cast_fp16")]; tensor var_3315_cast_fp16 = mul(x = current_key_29_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3315_cast_fp16")]; tensor key_57_cast_fp16 = add(x = var_3314_cast_fp16, y = var_3315_cast_fp16)[name = tensor("key_57_cast_fp16")]; tensor var_3318_cast_fp16 = mul(x = var_138_cast_fp16_14, y = var_239_cast_fp16)[name = tensor("op_3318_cast_fp16")]; tensor var_3319_cast_fp16 = mul(x = current_value_29_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3319_cast_fp16")]; tensor value_57_cast_fp16 = add(x = var_3318_cast_fp16, y = var_3319_cast_fp16)[name = tensor("value_57_cast_fp16")]; tensor var_3323 = const()[name = tensor("op_3323"), val = tensor([1, 20, 64, 1])]; tensor mh_q_57_cast_fp16 = reshape(shape = var_3323, x = query_57_cast_fp16)[name = tensor("mh_q_57_cast_fp16")]; tensor var_3325_to_fp16 = const()[name = tensor("op_3325_to_fp16"), val = tensor(0x1p-3)]; tensor var_3326_cast_fp16 = mul(x = mh_q_57_cast_fp16, y = var_3325_to_fp16)[name = tensor("op_3326_cast_fp16")]; tensor var_3329 = const()[name = tensor("op_3329"), val = tensor([1, 20, 64, 448])]; tensor var_3330_cast_fp16 = reshape(shape = var_3329, x = key_57_cast_fp16)[name = tensor("op_3330_cast_fp16")]; tensor mh_w_85_transpose_x_0 = const()[name = tensor("mh_w_85_transpose_x_0"), val = tensor(true)]; tensor mh_w_85_transpose_y_0 = const()[name = tensor("mh_w_85_transpose_y_0"), val = tensor(false)]; tensor mh_w_85_cast_fp16 = matmul(transpose_x = mh_w_85_transpose_x_0, transpose_y = mh_w_85_transpose_y_0, x = var_3326_cast_fp16, y = var_3330_cast_fp16)[name = tensor("mh_w_85_cast_fp16")]; tensor mh_w_87_cast_fp16 = add(x = mh_w_85_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_87_cast_fp16")]; tensor var_3338_cast_fp16 = softmax(axis = var_3250, x = mh_w_87_cast_fp16)[name = tensor("op_3338_cast_fp16")]; tensor var_3339 = const()[name = tensor("op_3339"), val = tensor([1, 20, 64, 448])]; tensor var_3340_cast_fp16 = reshape(shape = var_3339, x = value_57_cast_fp16)[name = tensor("op_3340_cast_fp16")]; tensor attn_57_transpose_x_0 = const()[name = tensor("attn_57_transpose_x_0"), val = tensor(false)]; tensor attn_57_transpose_y_0 = const()[name = tensor("attn_57_transpose_y_0"), val = tensor(true)]; tensor attn_57_cast_fp16 = matmul(transpose_x = attn_57_transpose_x_0, transpose_y = attn_57_transpose_y_0, x = var_3340_cast_fp16, y = var_3338_cast_fp16)[name = tensor("attn_57_cast_fp16")]; tensor var_3343 = const()[name = tensor("op_3343"), val = tensor([1, 1280, 1, 1])]; tensor input_141_cast_fp16 = reshape(shape = var_3343, x = attn_57_cast_fp16)[name = tensor("input_141_cast_fp16")]; tensor obj_203_pad_type_0 = const()[name = tensor("obj_203_pad_type_0"), val = tensor("valid")]; tensor obj_203_strides_0 = const()[name = tensor("obj_203_strides_0"), val = tensor([1, 1])]; tensor obj_203_pad_0 = const()[name = tensor("obj_203_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_203_dilations_0 = const()[name = tensor("obj_203_dilations_0"), val = tensor([1, 1])]; tensor obj_203_groups_0 = const()[name = tensor("obj_203_groups_0"), val = tensor(1)]; tensor layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(878401792)))]; tensor layers_14_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(881678656)))]; tensor obj_203_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_bias_to_fp16, dilations = obj_203_dilations_0, groups = obj_203_groups_0, pad = obj_203_pad_0, pad_type = obj_203_pad_type_0, strides = obj_203_strides_0, weight = layers_14_self_attn_o_proj_weight_to_fp16, x = input_141_cast_fp16)[name = tensor("obj_203_cast_fp16")]; tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = obj_203_cast_fp16)[name = tensor("inputs_87_cast_fp16")]; tensor out_87_axes_0 = const()[name = tensor("out_87_axes_0"), val = tensor([1])]; tensor var_3365_to_fp16 = const()[name = tensor("op_3365_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_3365_to_fp16, x = inputs_87_cast_fp16)[name = tensor("out_87_cast_fp16")]; tensor obj_205_gamma_0_to_fp16 = const()[name = tensor("obj_205_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(881681280)))]; tensor obj_205_beta_0_to_fp16 = const()[name = tensor("obj_205_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(881683904)))]; tensor obj_205_epsilon_0_to_fp16 = const()[name = tensor("obj_205_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_205_cast_fp16 = batch_norm(beta = obj_205_beta_0_to_fp16, epsilon = obj_205_epsilon_0_to_fp16, gamma = obj_205_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_87_cast_fp16)[name = tensor("obj_205_cast_fp16")]; tensor query_59_pad_type_0 = const()[name = tensor("query_59_pad_type_0"), val = tensor("valid")]; tensor query_59_strides_0 = const()[name = tensor("query_59_strides_0"), val = tensor([1, 1])]; tensor query_59_pad_0 = const()[name = tensor("query_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_59_dilations_0 = const()[name = tensor("query_59_dilations_0"), val = tensor([1, 1])]; tensor query_59_groups_0 = const()[name = tensor("query_59_groups_0"), val = tensor(1)]; tensor layers_14_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_14_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(881686528)))]; tensor layers_14_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_14_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(884963392)))]; tensor query_59_cast_fp16 = conv(bias = layers_14_encoder_attn_q_proj_bias_to_fp16, dilations = query_59_dilations_0, groups = query_59_groups_0, pad = query_59_pad_0, pad_type = query_59_pad_type_0, strides = query_59_strides_0, weight = layers_14_encoder_attn_q_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = tensor("query_59_cast_fp16")]; tensor key_59_pad_type_0 = const()[name = tensor("key_59_pad_type_0"), val = tensor("valid")]; tensor key_59_strides_0 = const()[name = tensor("key_59_strides_0"), val = tensor([1, 1])]; tensor key_59_pad_0 = const()[name = tensor("key_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_59_dilations_0 = const()[name = tensor("key_59_dilations_0"), val = tensor([1, 1])]; tensor key_59_groups_0 = const()[name = tensor("key_59_groups_0"), val = tensor(1)]; tensor layers_14_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_14_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(884966016)))]; tensor key_59_cast_fp16 = conv(dilations = key_59_dilations_0, groups = key_59_groups_0, pad = key_59_pad_0, pad_type = key_59_pad_type_0, strides = key_59_strides_0, weight = layers_14_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_59_cast_fp16")]; tensor value_59_pad_type_0 = const()[name = tensor("value_59_pad_type_0"), val = tensor("valid")]; tensor value_59_strides_0 = const()[name = tensor("value_59_strides_0"), val = tensor([1, 1])]; tensor value_59_pad_0 = const()[name = tensor("value_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_59_dilations_0 = const()[name = tensor("value_59_dilations_0"), val = tensor([1, 1])]; tensor value_59_groups_0 = const()[name = tensor("value_59_groups_0"), val = tensor(1)]; tensor layers_14_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_14_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(888242880)))]; tensor layers_14_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_14_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(891519744)))]; tensor value_59_cast_fp16 = conv(bias = layers_14_encoder_attn_v_proj_bias_to_fp16, dilations = value_59_dilations_0, groups = value_59_groups_0, pad = value_59_pad_0, pad_type = value_59_pad_type_0, strides = value_59_strides_0, weight = layers_14_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_59_cast_fp16")]; tensor var_3401 = const()[name = tensor("op_3401"), val = tensor([1, 20, 64, 1])]; tensor mh_q_59_cast_fp16 = reshape(shape = var_3401, x = query_59_cast_fp16)[name = tensor("mh_q_59_cast_fp16")]; tensor var_3403_to_fp16 = const()[name = tensor("op_3403_to_fp16"), val = tensor(0x1p-3)]; tensor var_3404_cast_fp16 = mul(x = mh_q_59_cast_fp16, y = var_3403_to_fp16)[name = tensor("op_3404_cast_fp16")]; tensor var_3407 = const()[name = tensor("op_3407"), val = tensor([1, 20, 64, 1500])]; tensor var_3408_cast_fp16 = reshape(shape = var_3407, x = key_59_cast_fp16)[name = tensor("op_3408_cast_fp16")]; tensor mh_w_89_transpose_x_0 = const()[name = tensor("mh_w_89_transpose_x_0"), val = tensor(true)]; tensor mh_w_89_transpose_y_0 = const()[name = tensor("mh_w_89_transpose_y_0"), val = tensor(false)]; tensor mh_w_89_cast_fp16 = matmul(transpose_x = mh_w_89_transpose_x_0, transpose_y = mh_w_89_transpose_y_0, x = var_3404_cast_fp16, y = var_3408_cast_fp16)[name = tensor("mh_w_89_cast_fp16")]; tensor obj_209_cast_fp16 = softmax(axis = var_3250, x = mh_w_89_cast_fp16)[name = tensor("obj_209_cast_fp16")]; tensor var_3412 = const()[name = tensor("op_3412"), val = tensor([1, 20, 64, 1500])]; tensor var_3413_cast_fp16 = reshape(shape = var_3412, x = value_59_cast_fp16)[name = tensor("op_3413_cast_fp16")]; tensor attn_59_transpose_x_0 = const()[name = tensor("attn_59_transpose_x_0"), val = tensor(false)]; tensor attn_59_transpose_y_0 = const()[name = tensor("attn_59_transpose_y_0"), val = tensor(true)]; tensor attn_59_cast_fp16 = matmul(transpose_x = attn_59_transpose_x_0, transpose_y = attn_59_transpose_y_0, x = var_3413_cast_fp16, y = obj_209_cast_fp16)[name = tensor("attn_59_cast_fp16")]; tensor var_3416 = const()[name = tensor("op_3416"), val = tensor([1, 1280, 1, 1])]; tensor input_143_cast_fp16 = reshape(shape = var_3416, x = attn_59_cast_fp16)[name = tensor("input_143_cast_fp16")]; tensor obj_207_pad_type_0 = const()[name = tensor("obj_207_pad_type_0"), val = tensor("valid")]; tensor obj_207_strides_0 = const()[name = tensor("obj_207_strides_0"), val = tensor([1, 1])]; tensor obj_207_pad_0 = const()[name = tensor("obj_207_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_207_dilations_0 = const()[name = tensor("obj_207_dilations_0"), val = tensor([1, 1])]; tensor obj_207_groups_0 = const()[name = tensor("obj_207_groups_0"), val = tensor(1)]; tensor layers_14_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_14_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(891522368)))]; tensor layers_14_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_14_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(894799232)))]; tensor obj_207_cast_fp16 = conv(bias = layers_14_encoder_attn_o_proj_bias_to_fp16, dilations = obj_207_dilations_0, groups = obj_207_groups_0, pad = obj_207_pad_0, pad_type = obj_207_pad_type_0, strides = obj_207_strides_0, weight = layers_14_encoder_attn_o_proj_weight_to_fp16, x = input_143_cast_fp16)[name = tensor("obj_207_cast_fp16")]; tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = obj_207_cast_fp16)[name = tensor("inputs_89_cast_fp16")]; tensor out_89_axes_0 = const()[name = tensor("out_89_axes_0"), val = tensor([1])]; tensor var_3434_to_fp16 = const()[name = tensor("op_3434_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_3434_to_fp16, x = inputs_89_cast_fp16)[name = tensor("out_89_cast_fp16")]; tensor input_145_gamma_0_to_fp16 = const()[name = tensor("input_145_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(894801856)))]; tensor input_145_beta_0_to_fp16 = const()[name = tensor("input_145_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(894804480)))]; tensor input_145_epsilon_0_to_fp16 = const()[name = tensor("input_145_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_145_cast_fp16 = batch_norm(beta = input_145_beta_0_to_fp16, epsilon = input_145_epsilon_0_to_fp16, gamma = input_145_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_89_cast_fp16)[name = tensor("input_145_cast_fp16")]; tensor input_147_pad_type_0 = const()[name = tensor("input_147_pad_type_0"), val = tensor("valid")]; tensor input_147_strides_0 = const()[name = tensor("input_147_strides_0"), val = tensor([1, 1])]; tensor input_147_pad_0 = const()[name = tensor("input_147_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_147_dilations_0 = const()[name = tensor("input_147_dilations_0"), val = tensor([1, 1])]; tensor input_147_groups_0 = const()[name = tensor("input_147_groups_0"), val = tensor(1)]; tensor layers_14_fc1_weight_to_fp16 = const()[name = tensor("layers_14_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(894807104)))]; tensor layers_14_fc1_bias_to_fp16 = const()[name = tensor("layers_14_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(907914368)))]; tensor input_147_cast_fp16 = conv(bias = layers_14_fc1_bias_to_fp16, dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = layers_14_fc1_weight_to_fp16, x = input_145_cast_fp16)[name = tensor("input_147_cast_fp16")]; tensor input_149_mode_0 = const()[name = tensor("input_149_mode_0"), val = tensor("EXACT")]; tensor input_149_cast_fp16 = gelu(mode = input_149_mode_0, x = input_147_cast_fp16)[name = tensor("input_149_cast_fp16")]; tensor hidden_states_31_pad_type_0 = const()[name = tensor("hidden_states_31_pad_type_0"), val = tensor("valid")]; tensor hidden_states_31_strides_0 = const()[name = tensor("hidden_states_31_strides_0"), val = tensor([1, 1])]; tensor hidden_states_31_pad_0 = const()[name = tensor("hidden_states_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_31_dilations_0 = const()[name = tensor("hidden_states_31_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_31_groups_0 = const()[name = tensor("hidden_states_31_groups_0"), val = tensor(1)]; tensor layers_14_fc2_weight_to_fp16 = const()[name = tensor("layers_14_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(907924672)))]; tensor layers_14_fc2_bias_to_fp16 = const()[name = tensor("layers_14_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(921031936)))]; tensor hidden_states_31_cast_fp16 = conv(bias = layers_14_fc2_bias_to_fp16, dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = layers_14_fc2_weight_to_fp16, x = input_149_cast_fp16)[name = tensor("hidden_states_31_cast_fp16")]; tensor inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = hidden_states_31_cast_fp16)[name = tensor("inputs_91_cast_fp16")]; tensor var_3469 = const()[name = tensor("op_3469"), val = tensor(3)]; tensor out_91_axes_0 = const()[name = tensor("out_91_axes_0"), val = tensor([1])]; tensor var_3494_to_fp16 = const()[name = tensor("op_3494_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_3494_to_fp16, x = inputs_91_cast_fp16)[name = tensor("out_91_cast_fp16")]; tensor obj_211_gamma_0_to_fp16 = const()[name = tensor("obj_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(921034560)))]; tensor obj_211_beta_0_to_fp16 = const()[name = tensor("obj_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(921037184)))]; tensor obj_211_epsilon_0_to_fp16 = const()[name = tensor("obj_211_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_211_cast_fp16 = batch_norm(beta = obj_211_beta_0_to_fp16, epsilon = obj_211_epsilon_0_to_fp16, gamma = obj_211_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_91_cast_fp16)[name = tensor("obj_211_cast_fp16")]; tensor query_61_pad_type_0 = const()[name = tensor("query_61_pad_type_0"), val = tensor("valid")]; tensor query_61_strides_0 = const()[name = tensor("query_61_strides_0"), val = tensor([1, 1])]; tensor query_61_pad_0 = const()[name = tensor("query_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_61_dilations_0 = const()[name = tensor("query_61_dilations_0"), val = tensor([1, 1])]; tensor query_61_groups_0 = const()[name = tensor("query_61_groups_0"), val = tensor(1)]; tensor layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(921039808)))]; tensor layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(924316672)))]; tensor query_61_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_bias_to_fp16, dilations = query_61_dilations_0, groups = query_61_groups_0, pad = query_61_pad_0, pad_type = query_61_pad_type_0, strides = query_61_strides_0, weight = layers_15_self_attn_q_proj_weight_to_fp16, x = obj_211_cast_fp16)[name = tensor("query_61_cast_fp16")]; tensor current_key_31_pad_type_0 = const()[name = tensor("current_key_31_pad_type_0"), val = tensor("valid")]; tensor current_key_31_strides_0 = const()[name = tensor("current_key_31_strides_0"), val = tensor([1, 1])]; tensor current_key_31_pad_0 = const()[name = tensor("current_key_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_31_dilations_0 = const()[name = tensor("current_key_31_dilations_0"), val = tensor([1, 1])]; tensor current_key_31_groups_0 = const()[name = tensor("current_key_31_groups_0"), val = tensor(1)]; tensor layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(924319296)))]; tensor current_key_31_cast_fp16 = conv(dilations = current_key_31_dilations_0, groups = current_key_31_groups_0, pad = current_key_31_pad_0, pad_type = current_key_31_pad_type_0, strides = current_key_31_strides_0, weight = layers_15_self_attn_k_proj_weight_to_fp16, x = obj_211_cast_fp16)[name = tensor("current_key_31_cast_fp16")]; tensor current_value_31_pad_type_0 = const()[name = tensor("current_value_31_pad_type_0"), val = tensor("valid")]; tensor current_value_31_strides_0 = const()[name = tensor("current_value_31_strides_0"), val = tensor([1, 1])]; tensor current_value_31_pad_0 = const()[name = tensor("current_value_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_31_dilations_0 = const()[name = tensor("current_value_31_dilations_0"), val = tensor([1, 1])]; tensor current_value_31_groups_0 = const()[name = tensor("current_value_31_groups_0"), val = tensor(1)]; tensor layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(927596160)))]; tensor layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(930873024)))]; tensor current_value_31_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_bias_to_fp16, dilations = current_value_31_dilations_0, groups = current_value_31_groups_0, pad = current_value_31_pad_0, pad_type = current_value_31_pad_type_0, strides = current_value_31_strides_0, weight = layers_15_self_attn_v_proj_weight_to_fp16, x = obj_211_cast_fp16)[name = tensor("current_value_31_cast_fp16")]; tensor var_3533_cast_fp16 = mul(x = var_103_cast_fp16_15, y = var_239_cast_fp16)[name = tensor("op_3533_cast_fp16")]; tensor var_3534_cast_fp16 = mul(x = current_key_31_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3534_cast_fp16")]; tensor key_61_cast_fp16 = add(x = var_3533_cast_fp16, y = var_3534_cast_fp16)[name = tensor("key_61_cast_fp16")]; tensor var_3537_cast_fp16 = mul(x = var_138_cast_fp16_15, y = var_239_cast_fp16)[name = tensor("op_3537_cast_fp16")]; tensor var_3538_cast_fp16 = mul(x = current_value_31_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3538_cast_fp16")]; tensor value_61_cast_fp16 = add(x = var_3537_cast_fp16, y = var_3538_cast_fp16)[name = tensor("value_61_cast_fp16")]; tensor var_3542 = const()[name = tensor("op_3542"), val = tensor([1, 20, 64, 1])]; tensor mh_q_61_cast_fp16 = reshape(shape = var_3542, x = query_61_cast_fp16)[name = tensor("mh_q_61_cast_fp16")]; tensor var_3544_to_fp16 = const()[name = tensor("op_3544_to_fp16"), val = tensor(0x1p-3)]; tensor var_3545_cast_fp16 = mul(x = mh_q_61_cast_fp16, y = var_3544_to_fp16)[name = tensor("op_3545_cast_fp16")]; tensor var_3548 = const()[name = tensor("op_3548"), val = tensor([1, 20, 64, 448])]; tensor var_3549_cast_fp16 = reshape(shape = var_3548, x = key_61_cast_fp16)[name = tensor("op_3549_cast_fp16")]; tensor mh_w_91_transpose_x_0 = const()[name = tensor("mh_w_91_transpose_x_0"), val = tensor(true)]; tensor mh_w_91_transpose_y_0 = const()[name = tensor("mh_w_91_transpose_y_0"), val = tensor(false)]; tensor mh_w_91_cast_fp16 = matmul(transpose_x = mh_w_91_transpose_x_0, transpose_y = mh_w_91_transpose_y_0, x = var_3545_cast_fp16, y = var_3549_cast_fp16)[name = tensor("mh_w_91_cast_fp16")]; tensor mh_w_93_cast_fp16 = add(x = mh_w_91_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_93_cast_fp16")]; tensor var_3557_cast_fp16 = softmax(axis = var_3469, x = mh_w_93_cast_fp16)[name = tensor("op_3557_cast_fp16")]; tensor var_3558 = const()[name = tensor("op_3558"), val = tensor([1, 20, 64, 448])]; tensor var_3559_cast_fp16 = reshape(shape = var_3558, x = value_61_cast_fp16)[name = tensor("op_3559_cast_fp16")]; tensor attn_61_transpose_x_0 = const()[name = tensor("attn_61_transpose_x_0"), val = tensor(false)]; tensor attn_61_transpose_y_0 = const()[name = tensor("attn_61_transpose_y_0"), val = tensor(true)]; tensor attn_61_cast_fp16 = matmul(transpose_x = attn_61_transpose_x_0, transpose_y = attn_61_transpose_y_0, x = var_3559_cast_fp16, y = var_3557_cast_fp16)[name = tensor("attn_61_cast_fp16")]; tensor var_3562 = const()[name = tensor("op_3562"), val = tensor([1, 1280, 1, 1])]; tensor input_151_cast_fp16 = reshape(shape = var_3562, x = attn_61_cast_fp16)[name = tensor("input_151_cast_fp16")]; tensor obj_217_pad_type_0 = const()[name = tensor("obj_217_pad_type_0"), val = tensor("valid")]; tensor obj_217_strides_0 = const()[name = tensor("obj_217_strides_0"), val = tensor([1, 1])]; tensor obj_217_pad_0 = const()[name = tensor("obj_217_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_217_dilations_0 = const()[name = tensor("obj_217_dilations_0"), val = tensor([1, 1])]; tensor obj_217_groups_0 = const()[name = tensor("obj_217_groups_0"), val = tensor(1)]; tensor layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(930875648)))]; tensor layers_15_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(934152512)))]; tensor obj_217_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_bias_to_fp16, dilations = obj_217_dilations_0, groups = obj_217_groups_0, pad = obj_217_pad_0, pad_type = obj_217_pad_type_0, strides = obj_217_strides_0, weight = layers_15_self_attn_o_proj_weight_to_fp16, x = input_151_cast_fp16)[name = tensor("obj_217_cast_fp16")]; tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = obj_217_cast_fp16)[name = tensor("inputs_93_cast_fp16")]; tensor out_93_axes_0 = const()[name = tensor("out_93_axes_0"), val = tensor([1])]; tensor var_3584_to_fp16 = const()[name = tensor("op_3584_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_3584_to_fp16, x = inputs_93_cast_fp16)[name = tensor("out_93_cast_fp16")]; tensor obj_219_gamma_0_to_fp16 = const()[name = tensor("obj_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(934155136)))]; tensor obj_219_beta_0_to_fp16 = const()[name = tensor("obj_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(934157760)))]; tensor obj_219_epsilon_0_to_fp16 = const()[name = tensor("obj_219_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_219_cast_fp16 = batch_norm(beta = obj_219_beta_0_to_fp16, epsilon = obj_219_epsilon_0_to_fp16, gamma = obj_219_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_93_cast_fp16)[name = tensor("obj_219_cast_fp16")]; tensor query_63_pad_type_0 = const()[name = tensor("query_63_pad_type_0"), val = tensor("valid")]; tensor query_63_strides_0 = const()[name = tensor("query_63_strides_0"), val = tensor([1, 1])]; tensor query_63_pad_0 = const()[name = tensor("query_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_63_dilations_0 = const()[name = tensor("query_63_dilations_0"), val = tensor([1, 1])]; tensor query_63_groups_0 = const()[name = tensor("query_63_groups_0"), val = tensor(1)]; tensor layers_15_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_15_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(934160384)))]; tensor layers_15_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_15_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(937437248)))]; tensor query_63_cast_fp16 = conv(bias = layers_15_encoder_attn_q_proj_bias_to_fp16, dilations = query_63_dilations_0, groups = query_63_groups_0, pad = query_63_pad_0, pad_type = query_63_pad_type_0, strides = query_63_strides_0, weight = layers_15_encoder_attn_q_proj_weight_to_fp16, x = obj_219_cast_fp16)[name = tensor("query_63_cast_fp16")]; tensor key_63_pad_type_0 = const()[name = tensor("key_63_pad_type_0"), val = tensor("valid")]; tensor key_63_strides_0 = const()[name = tensor("key_63_strides_0"), val = tensor([1, 1])]; tensor key_63_pad_0 = const()[name = tensor("key_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_63_dilations_0 = const()[name = tensor("key_63_dilations_0"), val = tensor([1, 1])]; tensor key_63_groups_0 = const()[name = tensor("key_63_groups_0"), val = tensor(1)]; tensor layers_15_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_15_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(937439872)))]; tensor key_63_cast_fp16 = conv(dilations = key_63_dilations_0, groups = key_63_groups_0, pad = key_63_pad_0, pad_type = key_63_pad_type_0, strides = key_63_strides_0, weight = layers_15_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_63_cast_fp16")]; tensor value_63_pad_type_0 = const()[name = tensor("value_63_pad_type_0"), val = tensor("valid")]; tensor value_63_strides_0 = const()[name = tensor("value_63_strides_0"), val = tensor([1, 1])]; tensor value_63_pad_0 = const()[name = tensor("value_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_63_dilations_0 = const()[name = tensor("value_63_dilations_0"), val = tensor([1, 1])]; tensor value_63_groups_0 = const()[name = tensor("value_63_groups_0"), val = tensor(1)]; tensor layers_15_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_15_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(940716736)))]; tensor layers_15_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_15_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(943993600)))]; tensor value_63_cast_fp16 = conv(bias = layers_15_encoder_attn_v_proj_bias_to_fp16, dilations = value_63_dilations_0, groups = value_63_groups_0, pad = value_63_pad_0, pad_type = value_63_pad_type_0, strides = value_63_strides_0, weight = layers_15_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_63_cast_fp16")]; tensor var_3620 = const()[name = tensor("op_3620"), val = tensor([1, 20, 64, 1])]; tensor mh_q_63_cast_fp16 = reshape(shape = var_3620, x = query_63_cast_fp16)[name = tensor("mh_q_63_cast_fp16")]; tensor var_3622_to_fp16 = const()[name = tensor("op_3622_to_fp16"), val = tensor(0x1p-3)]; tensor var_3623_cast_fp16 = mul(x = mh_q_63_cast_fp16, y = var_3622_to_fp16)[name = tensor("op_3623_cast_fp16")]; tensor var_3626 = const()[name = tensor("op_3626"), val = tensor([1, 20, 64, 1500])]; tensor var_3627_cast_fp16 = reshape(shape = var_3626, x = key_63_cast_fp16)[name = tensor("op_3627_cast_fp16")]; tensor mh_w_95_transpose_x_0 = const()[name = tensor("mh_w_95_transpose_x_0"), val = tensor(true)]; tensor mh_w_95_transpose_y_0 = const()[name = tensor("mh_w_95_transpose_y_0"), val = tensor(false)]; tensor mh_w_95_cast_fp16 = matmul(transpose_x = mh_w_95_transpose_x_0, transpose_y = mh_w_95_transpose_y_0, x = var_3623_cast_fp16, y = var_3627_cast_fp16)[name = tensor("mh_w_95_cast_fp16")]; tensor obj_223_cast_fp16 = softmax(axis = var_3469, x = mh_w_95_cast_fp16)[name = tensor("obj_223_cast_fp16")]; tensor var_3631 = const()[name = tensor("op_3631"), val = tensor([1, 20, 64, 1500])]; tensor var_3632_cast_fp16 = reshape(shape = var_3631, x = value_63_cast_fp16)[name = tensor("op_3632_cast_fp16")]; tensor attn_63_transpose_x_0 = const()[name = tensor("attn_63_transpose_x_0"), val = tensor(false)]; tensor attn_63_transpose_y_0 = const()[name = tensor("attn_63_transpose_y_0"), val = tensor(true)]; tensor attn_63_cast_fp16 = matmul(transpose_x = attn_63_transpose_x_0, transpose_y = attn_63_transpose_y_0, x = var_3632_cast_fp16, y = obj_223_cast_fp16)[name = tensor("attn_63_cast_fp16")]; tensor var_3635 = const()[name = tensor("op_3635"), val = tensor([1, 1280, 1, 1])]; tensor input_153_cast_fp16 = reshape(shape = var_3635, x = attn_63_cast_fp16)[name = tensor("input_153_cast_fp16")]; tensor obj_221_pad_type_0 = const()[name = tensor("obj_221_pad_type_0"), val = tensor("valid")]; tensor obj_221_strides_0 = const()[name = tensor("obj_221_strides_0"), val = tensor([1, 1])]; tensor obj_221_pad_0 = const()[name = tensor("obj_221_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_221_dilations_0 = const()[name = tensor("obj_221_dilations_0"), val = tensor([1, 1])]; tensor obj_221_groups_0 = const()[name = tensor("obj_221_groups_0"), val = tensor(1)]; tensor layers_15_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_15_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(943996224)))]; tensor layers_15_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_15_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(947273088)))]; tensor obj_221_cast_fp16 = conv(bias = layers_15_encoder_attn_o_proj_bias_to_fp16, dilations = obj_221_dilations_0, groups = obj_221_groups_0, pad = obj_221_pad_0, pad_type = obj_221_pad_type_0, strides = obj_221_strides_0, weight = layers_15_encoder_attn_o_proj_weight_to_fp16, x = input_153_cast_fp16)[name = tensor("obj_221_cast_fp16")]; tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_221_cast_fp16)[name = tensor("inputs_95_cast_fp16")]; tensor out_95_axes_0 = const()[name = tensor("out_95_axes_0"), val = tensor([1])]; tensor var_3653_to_fp16 = const()[name = tensor("op_3653_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_3653_to_fp16, x = inputs_95_cast_fp16)[name = tensor("out_95_cast_fp16")]; tensor input_155_gamma_0_to_fp16 = const()[name = tensor("input_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(947275712)))]; tensor input_155_beta_0_to_fp16 = const()[name = tensor("input_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(947278336)))]; tensor input_155_epsilon_0_to_fp16 = const()[name = tensor("input_155_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_95_cast_fp16)[name = tensor("input_155_cast_fp16")]; tensor input_157_pad_type_0 = const()[name = tensor("input_157_pad_type_0"), val = tensor("valid")]; tensor input_157_strides_0 = const()[name = tensor("input_157_strides_0"), val = tensor([1, 1])]; tensor input_157_pad_0 = const()[name = tensor("input_157_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_157_dilations_0 = const()[name = tensor("input_157_dilations_0"), val = tensor([1, 1])]; tensor input_157_groups_0 = const()[name = tensor("input_157_groups_0"), val = tensor(1)]; tensor layers_15_fc1_weight_to_fp16 = const()[name = tensor("layers_15_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(947280960)))]; tensor layers_15_fc1_bias_to_fp16 = const()[name = tensor("layers_15_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(960388224)))]; tensor input_157_cast_fp16 = conv(bias = layers_15_fc1_bias_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = layers_15_fc1_weight_to_fp16, x = input_155_cast_fp16)[name = tensor("input_157_cast_fp16")]; tensor input_159_mode_0 = const()[name = tensor("input_159_mode_0"), val = tensor("EXACT")]; tensor input_159_cast_fp16 = gelu(mode = input_159_mode_0, x = input_157_cast_fp16)[name = tensor("input_159_cast_fp16")]; tensor hidden_states_33_pad_type_0 = const()[name = tensor("hidden_states_33_pad_type_0"), val = tensor("valid")]; tensor hidden_states_33_strides_0 = const()[name = tensor("hidden_states_33_strides_0"), val = tensor([1, 1])]; tensor hidden_states_33_pad_0 = const()[name = tensor("hidden_states_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_33_dilations_0 = const()[name = tensor("hidden_states_33_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_33_groups_0 = const()[name = tensor("hidden_states_33_groups_0"), val = tensor(1)]; tensor layers_15_fc2_weight_to_fp16 = const()[name = tensor("layers_15_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(960398528)))]; tensor layers_15_fc2_bias_to_fp16 = const()[name = tensor("layers_15_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(973505792)))]; tensor hidden_states_33_cast_fp16 = conv(bias = layers_15_fc2_bias_to_fp16, dilations = hidden_states_33_dilations_0, groups = hidden_states_33_groups_0, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = hidden_states_33_strides_0, weight = layers_15_fc2_weight_to_fp16, x = input_159_cast_fp16)[name = tensor("hidden_states_33_cast_fp16")]; tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = hidden_states_33_cast_fp16)[name = tensor("inputs_97_cast_fp16")]; tensor var_3688 = const()[name = tensor("op_3688"), val = tensor(3)]; tensor out_97_axes_0 = const()[name = tensor("out_97_axes_0"), val = tensor([1])]; tensor var_3713_to_fp16 = const()[name = tensor("op_3713_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_3713_to_fp16, x = inputs_97_cast_fp16)[name = tensor("out_97_cast_fp16")]; tensor obj_225_gamma_0_to_fp16 = const()[name = tensor("obj_225_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(973508416)))]; tensor obj_225_beta_0_to_fp16 = const()[name = tensor("obj_225_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(973511040)))]; tensor obj_225_epsilon_0_to_fp16 = const()[name = tensor("obj_225_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_225_cast_fp16 = batch_norm(beta = obj_225_beta_0_to_fp16, epsilon = obj_225_epsilon_0_to_fp16, gamma = obj_225_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_97_cast_fp16)[name = tensor("obj_225_cast_fp16")]; tensor query_65_pad_type_0 = const()[name = tensor("query_65_pad_type_0"), val = tensor("valid")]; tensor query_65_strides_0 = const()[name = tensor("query_65_strides_0"), val = tensor([1, 1])]; tensor query_65_pad_0 = const()[name = tensor("query_65_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_65_dilations_0 = const()[name = tensor("query_65_dilations_0"), val = tensor([1, 1])]; tensor query_65_groups_0 = const()[name = tensor("query_65_groups_0"), val = tensor(1)]; tensor layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(973513664)))]; tensor layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(976790528)))]; tensor query_65_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_bias_to_fp16, dilations = query_65_dilations_0, groups = query_65_groups_0, pad = query_65_pad_0, pad_type = query_65_pad_type_0, strides = query_65_strides_0, weight = layers_16_self_attn_q_proj_weight_to_fp16, x = obj_225_cast_fp16)[name = tensor("query_65_cast_fp16")]; tensor current_key_33_pad_type_0 = const()[name = tensor("current_key_33_pad_type_0"), val = tensor("valid")]; tensor current_key_33_strides_0 = const()[name = tensor("current_key_33_strides_0"), val = tensor([1, 1])]; tensor current_key_33_pad_0 = const()[name = tensor("current_key_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_33_dilations_0 = const()[name = tensor("current_key_33_dilations_0"), val = tensor([1, 1])]; tensor current_key_33_groups_0 = const()[name = tensor("current_key_33_groups_0"), val = tensor(1)]; tensor layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(976793152)))]; tensor current_key_33_cast_fp16 = conv(dilations = current_key_33_dilations_0, groups = current_key_33_groups_0, pad = current_key_33_pad_0, pad_type = current_key_33_pad_type_0, strides = current_key_33_strides_0, weight = layers_16_self_attn_k_proj_weight_to_fp16, x = obj_225_cast_fp16)[name = tensor("current_key_33_cast_fp16")]; tensor current_value_33_pad_type_0 = const()[name = tensor("current_value_33_pad_type_0"), val = tensor("valid")]; tensor current_value_33_strides_0 = const()[name = tensor("current_value_33_strides_0"), val = tensor([1, 1])]; tensor current_value_33_pad_0 = const()[name = tensor("current_value_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_33_dilations_0 = const()[name = tensor("current_value_33_dilations_0"), val = tensor([1, 1])]; tensor current_value_33_groups_0 = const()[name = tensor("current_value_33_groups_0"), val = tensor(1)]; tensor layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(980070016)))]; tensor layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(983346880)))]; tensor current_value_33_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_bias_to_fp16, dilations = current_value_33_dilations_0, groups = current_value_33_groups_0, pad = current_value_33_pad_0, pad_type = current_value_33_pad_type_0, strides = current_value_33_strides_0, weight = layers_16_self_attn_v_proj_weight_to_fp16, x = obj_225_cast_fp16)[name = tensor("current_value_33_cast_fp16")]; tensor var_3752_cast_fp16 = mul(x = var_103_cast_fp16_16, y = var_239_cast_fp16)[name = tensor("op_3752_cast_fp16")]; tensor var_3753_cast_fp16 = mul(x = current_key_33_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3753_cast_fp16")]; tensor key_65_cast_fp16 = add(x = var_3752_cast_fp16, y = var_3753_cast_fp16)[name = tensor("key_65_cast_fp16")]; tensor var_3756_cast_fp16 = mul(x = var_138_cast_fp16_16, y = var_239_cast_fp16)[name = tensor("op_3756_cast_fp16")]; tensor var_3757_cast_fp16 = mul(x = current_value_33_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3757_cast_fp16")]; tensor value_65_cast_fp16 = add(x = var_3756_cast_fp16, y = var_3757_cast_fp16)[name = tensor("value_65_cast_fp16")]; tensor var_3761 = const()[name = tensor("op_3761"), val = tensor([1, 20, 64, 1])]; tensor mh_q_65_cast_fp16 = reshape(shape = var_3761, x = query_65_cast_fp16)[name = tensor("mh_q_65_cast_fp16")]; tensor var_3763_to_fp16 = const()[name = tensor("op_3763_to_fp16"), val = tensor(0x1p-3)]; tensor var_3764_cast_fp16 = mul(x = mh_q_65_cast_fp16, y = var_3763_to_fp16)[name = tensor("op_3764_cast_fp16")]; tensor var_3767 = const()[name = tensor("op_3767"), val = tensor([1, 20, 64, 448])]; tensor var_3768_cast_fp16 = reshape(shape = var_3767, x = key_65_cast_fp16)[name = tensor("op_3768_cast_fp16")]; tensor mh_w_97_transpose_x_0 = const()[name = tensor("mh_w_97_transpose_x_0"), val = tensor(true)]; tensor mh_w_97_transpose_y_0 = const()[name = tensor("mh_w_97_transpose_y_0"), val = tensor(false)]; tensor mh_w_97_cast_fp16 = matmul(transpose_x = mh_w_97_transpose_x_0, transpose_y = mh_w_97_transpose_y_0, x = var_3764_cast_fp16, y = var_3768_cast_fp16)[name = tensor("mh_w_97_cast_fp16")]; tensor mh_w_99_cast_fp16 = add(x = mh_w_97_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_99_cast_fp16")]; tensor var_3776_cast_fp16 = softmax(axis = var_3688, x = mh_w_99_cast_fp16)[name = tensor("op_3776_cast_fp16")]; tensor var_3777 = const()[name = tensor("op_3777"), val = tensor([1, 20, 64, 448])]; tensor var_3778_cast_fp16 = reshape(shape = var_3777, x = value_65_cast_fp16)[name = tensor("op_3778_cast_fp16")]; tensor attn_65_transpose_x_0 = const()[name = tensor("attn_65_transpose_x_0"), val = tensor(false)]; tensor attn_65_transpose_y_0 = const()[name = tensor("attn_65_transpose_y_0"), val = tensor(true)]; tensor attn_65_cast_fp16 = matmul(transpose_x = attn_65_transpose_x_0, transpose_y = attn_65_transpose_y_0, x = var_3778_cast_fp16, y = var_3776_cast_fp16)[name = tensor("attn_65_cast_fp16")]; tensor var_3781 = const()[name = tensor("op_3781"), val = tensor([1, 1280, 1, 1])]; tensor input_161_cast_fp16 = reshape(shape = var_3781, x = attn_65_cast_fp16)[name = tensor("input_161_cast_fp16")]; tensor obj_231_pad_type_0 = const()[name = tensor("obj_231_pad_type_0"), val = tensor("valid")]; tensor obj_231_strides_0 = const()[name = tensor("obj_231_strides_0"), val = tensor([1, 1])]; tensor obj_231_pad_0 = const()[name = tensor("obj_231_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_231_dilations_0 = const()[name = tensor("obj_231_dilations_0"), val = tensor([1, 1])]; tensor obj_231_groups_0 = const()[name = tensor("obj_231_groups_0"), val = tensor(1)]; tensor layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(983349504)))]; tensor layers_16_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(986626368)))]; tensor obj_231_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_bias_to_fp16, dilations = obj_231_dilations_0, groups = obj_231_groups_0, pad = obj_231_pad_0, pad_type = obj_231_pad_type_0, strides = obj_231_strides_0, weight = layers_16_self_attn_o_proj_weight_to_fp16, x = input_161_cast_fp16)[name = tensor("obj_231_cast_fp16")]; tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = obj_231_cast_fp16)[name = tensor("inputs_99_cast_fp16")]; tensor out_99_axes_0 = const()[name = tensor("out_99_axes_0"), val = tensor([1])]; tensor var_3803_to_fp16 = const()[name = tensor("op_3803_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_3803_to_fp16, x = inputs_99_cast_fp16)[name = tensor("out_99_cast_fp16")]; tensor obj_233_gamma_0_to_fp16 = const()[name = tensor("obj_233_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(986628992)))]; tensor obj_233_beta_0_to_fp16 = const()[name = tensor("obj_233_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(986631616)))]; tensor obj_233_epsilon_0_to_fp16 = const()[name = tensor("obj_233_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_233_cast_fp16 = batch_norm(beta = obj_233_beta_0_to_fp16, epsilon = obj_233_epsilon_0_to_fp16, gamma = obj_233_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_99_cast_fp16)[name = tensor("obj_233_cast_fp16")]; tensor query_67_pad_type_0 = const()[name = tensor("query_67_pad_type_0"), val = tensor("valid")]; tensor query_67_strides_0 = const()[name = tensor("query_67_strides_0"), val = tensor([1, 1])]; tensor query_67_pad_0 = const()[name = tensor("query_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_67_dilations_0 = const()[name = tensor("query_67_dilations_0"), val = tensor([1, 1])]; tensor query_67_groups_0 = const()[name = tensor("query_67_groups_0"), val = tensor(1)]; tensor layers_16_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_16_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(986634240)))]; tensor layers_16_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_16_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(989911104)))]; tensor query_67_cast_fp16 = conv(bias = layers_16_encoder_attn_q_proj_bias_to_fp16, dilations = query_67_dilations_0, groups = query_67_groups_0, pad = query_67_pad_0, pad_type = query_67_pad_type_0, strides = query_67_strides_0, weight = layers_16_encoder_attn_q_proj_weight_to_fp16, x = obj_233_cast_fp16)[name = tensor("query_67_cast_fp16")]; tensor key_67_pad_type_0 = const()[name = tensor("key_67_pad_type_0"), val = tensor("valid")]; tensor key_67_strides_0 = const()[name = tensor("key_67_strides_0"), val = tensor([1, 1])]; tensor key_67_pad_0 = const()[name = tensor("key_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_67_dilations_0 = const()[name = tensor("key_67_dilations_0"), val = tensor([1, 1])]; tensor key_67_groups_0 = const()[name = tensor("key_67_groups_0"), val = tensor(1)]; tensor layers_16_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_16_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(989913728)))]; tensor key_67_cast_fp16 = conv(dilations = key_67_dilations_0, groups = key_67_groups_0, pad = key_67_pad_0, pad_type = key_67_pad_type_0, strides = key_67_strides_0, weight = layers_16_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_67_cast_fp16")]; tensor value_67_pad_type_0 = const()[name = tensor("value_67_pad_type_0"), val = tensor("valid")]; tensor value_67_strides_0 = const()[name = tensor("value_67_strides_0"), val = tensor([1, 1])]; tensor value_67_pad_0 = const()[name = tensor("value_67_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_67_dilations_0 = const()[name = tensor("value_67_dilations_0"), val = tensor([1, 1])]; tensor value_67_groups_0 = const()[name = tensor("value_67_groups_0"), val = tensor(1)]; tensor layers_16_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_16_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(993190592)))]; tensor layers_16_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_16_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(996467456)))]; tensor value_67_cast_fp16 = conv(bias = layers_16_encoder_attn_v_proj_bias_to_fp16, dilations = value_67_dilations_0, groups = value_67_groups_0, pad = value_67_pad_0, pad_type = value_67_pad_type_0, strides = value_67_strides_0, weight = layers_16_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_67_cast_fp16")]; tensor var_3839 = const()[name = tensor("op_3839"), val = tensor([1, 20, 64, 1])]; tensor mh_q_67_cast_fp16 = reshape(shape = var_3839, x = query_67_cast_fp16)[name = tensor("mh_q_67_cast_fp16")]; tensor var_3841_to_fp16 = const()[name = tensor("op_3841_to_fp16"), val = tensor(0x1p-3)]; tensor var_3842_cast_fp16 = mul(x = mh_q_67_cast_fp16, y = var_3841_to_fp16)[name = tensor("op_3842_cast_fp16")]; tensor var_3845 = const()[name = tensor("op_3845"), val = tensor([1, 20, 64, 1500])]; tensor var_3846_cast_fp16 = reshape(shape = var_3845, x = key_67_cast_fp16)[name = tensor("op_3846_cast_fp16")]; tensor mh_w_101_transpose_x_0 = const()[name = tensor("mh_w_101_transpose_x_0"), val = tensor(true)]; tensor mh_w_101_transpose_y_0 = const()[name = tensor("mh_w_101_transpose_y_0"), val = tensor(false)]; tensor mh_w_101_cast_fp16 = matmul(transpose_x = mh_w_101_transpose_x_0, transpose_y = mh_w_101_transpose_y_0, x = var_3842_cast_fp16, y = var_3846_cast_fp16)[name = tensor("mh_w_101_cast_fp16")]; tensor obj_237_cast_fp16 = softmax(axis = var_3688, x = mh_w_101_cast_fp16)[name = tensor("obj_237_cast_fp16")]; tensor var_3850 = const()[name = tensor("op_3850"), val = tensor([1, 20, 64, 1500])]; tensor var_3851_cast_fp16 = reshape(shape = var_3850, x = value_67_cast_fp16)[name = tensor("op_3851_cast_fp16")]; tensor attn_67_transpose_x_0 = const()[name = tensor("attn_67_transpose_x_0"), val = tensor(false)]; tensor attn_67_transpose_y_0 = const()[name = tensor("attn_67_transpose_y_0"), val = tensor(true)]; tensor attn_67_cast_fp16 = matmul(transpose_x = attn_67_transpose_x_0, transpose_y = attn_67_transpose_y_0, x = var_3851_cast_fp16, y = obj_237_cast_fp16)[name = tensor("attn_67_cast_fp16")]; tensor var_3854 = const()[name = tensor("op_3854"), val = tensor([1, 1280, 1, 1])]; tensor input_163_cast_fp16 = reshape(shape = var_3854, x = attn_67_cast_fp16)[name = tensor("input_163_cast_fp16")]; tensor obj_235_pad_type_0 = const()[name = tensor("obj_235_pad_type_0"), val = tensor("valid")]; tensor obj_235_strides_0 = const()[name = tensor("obj_235_strides_0"), val = tensor([1, 1])]; tensor obj_235_pad_0 = const()[name = tensor("obj_235_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_235_dilations_0 = const()[name = tensor("obj_235_dilations_0"), val = tensor([1, 1])]; tensor obj_235_groups_0 = const()[name = tensor("obj_235_groups_0"), val = tensor(1)]; tensor layers_16_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_16_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(996470080)))]; tensor layers_16_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_16_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(999746944)))]; tensor obj_235_cast_fp16 = conv(bias = layers_16_encoder_attn_o_proj_bias_to_fp16, dilations = obj_235_dilations_0, groups = obj_235_groups_0, pad = obj_235_pad_0, pad_type = obj_235_pad_type_0, strides = obj_235_strides_0, weight = layers_16_encoder_attn_o_proj_weight_to_fp16, x = input_163_cast_fp16)[name = tensor("obj_235_cast_fp16")]; tensor inputs_101_cast_fp16 = add(x = inputs_99_cast_fp16, y = obj_235_cast_fp16)[name = tensor("inputs_101_cast_fp16")]; tensor out_101_axes_0 = const()[name = tensor("out_101_axes_0"), val = tensor([1])]; tensor var_3875_to_fp16 = const()[name = tensor("op_3875_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_3875_to_fp16, x = inputs_101_cast_fp16)[name = tensor("out_101_cast_fp16")]; tensor input_165_gamma_0_to_fp16 = const()[name = tensor("input_165_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(999749568)))]; tensor input_165_beta_0_to_fp16 = const()[name = tensor("input_165_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(999752192)))]; tensor input_165_epsilon_0_to_fp16 = const()[name = tensor("input_165_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_165_cast_fp16 = batch_norm(beta = input_165_beta_0_to_fp16, epsilon = input_165_epsilon_0_to_fp16, gamma = input_165_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_101_cast_fp16)[name = tensor("input_165_cast_fp16")]; tensor input_167_pad_type_0 = const()[name = tensor("input_167_pad_type_0"), val = tensor("valid")]; tensor input_167_strides_0 = const()[name = tensor("input_167_strides_0"), val = tensor([1, 1])]; tensor input_167_pad_0 = const()[name = tensor("input_167_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_167_dilations_0 = const()[name = tensor("input_167_dilations_0"), val = tensor([1, 1])]; tensor input_167_groups_0 = const()[name = tensor("input_167_groups_0"), val = tensor(1)]; tensor layers_16_fc1_weight_to_fp16 = const()[name = tensor("layers_16_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(999754816)))]; tensor layers_16_fc1_bias_to_fp16 = const()[name = tensor("layers_16_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1012862080)))]; tensor input_167_cast_fp16 = conv(bias = layers_16_fc1_bias_to_fp16, dilations = input_167_dilations_0, groups = input_167_groups_0, pad = input_167_pad_0, pad_type = input_167_pad_type_0, strides = input_167_strides_0, weight = layers_16_fc1_weight_to_fp16, x = input_165_cast_fp16)[name = tensor("input_167_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 = input_167_cast_fp16)[name = tensor("input_169_cast_fp16")]; tensor hidden_states_35_pad_type_0 = const()[name = tensor("hidden_states_35_pad_type_0"), val = tensor("valid")]; tensor hidden_states_35_strides_0 = const()[name = tensor("hidden_states_35_strides_0"), val = tensor([1, 1])]; tensor hidden_states_35_pad_0 = const()[name = tensor("hidden_states_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_35_dilations_0 = const()[name = tensor("hidden_states_35_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_35_groups_0 = const()[name = tensor("hidden_states_35_groups_0"), val = tensor(1)]; tensor layers_16_fc2_weight_to_fp16 = const()[name = tensor("layers_16_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1012872384)))]; tensor layers_16_fc2_bias_to_fp16 = const()[name = tensor("layers_16_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1025979648)))]; tensor hidden_states_35_cast_fp16 = conv(bias = layers_16_fc2_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = layers_16_fc2_weight_to_fp16, x = input_169_cast_fp16)[name = tensor("hidden_states_35_cast_fp16")]; tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = hidden_states_35_cast_fp16)[name = tensor("inputs_103_cast_fp16")]; tensor var_3911 = const()[name = tensor("op_3911"), val = tensor(3)]; tensor out_103_axes_0 = const()[name = tensor("out_103_axes_0"), val = tensor([1])]; tensor var_3936_to_fp16 = const()[name = tensor("op_3936_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_3936_to_fp16, x = inputs_103_cast_fp16)[name = tensor("out_103_cast_fp16")]; tensor obj_239_gamma_0_to_fp16 = const()[name = tensor("obj_239_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1025982272)))]; tensor obj_239_beta_0_to_fp16 = const()[name = tensor("obj_239_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1025984896)))]; tensor obj_239_epsilon_0_to_fp16 = const()[name = tensor("obj_239_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_239_cast_fp16 = batch_norm(beta = obj_239_beta_0_to_fp16, epsilon = obj_239_epsilon_0_to_fp16, gamma = obj_239_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_103_cast_fp16)[name = tensor("obj_239_cast_fp16")]; tensor query_69_pad_type_0 = const()[name = tensor("query_69_pad_type_0"), val = tensor("valid")]; tensor query_69_strides_0 = const()[name = tensor("query_69_strides_0"), val = tensor([1, 1])]; tensor query_69_pad_0 = const()[name = tensor("query_69_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_69_dilations_0 = const()[name = tensor("query_69_dilations_0"), val = tensor([1, 1])]; tensor query_69_groups_0 = const()[name = tensor("query_69_groups_0"), val = tensor(1)]; tensor layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1025987520)))]; tensor layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1029264384)))]; tensor query_69_cast_fp16 = conv(bias = layers_17_self_attn_q_proj_bias_to_fp16, dilations = query_69_dilations_0, groups = query_69_groups_0, pad = query_69_pad_0, pad_type = query_69_pad_type_0, strides = query_69_strides_0, weight = layers_17_self_attn_q_proj_weight_to_fp16, x = obj_239_cast_fp16)[name = tensor("query_69_cast_fp16")]; tensor current_key_35_pad_type_0 = const()[name = tensor("current_key_35_pad_type_0"), val = tensor("valid")]; tensor current_key_35_strides_0 = const()[name = tensor("current_key_35_strides_0"), val = tensor([1, 1])]; tensor current_key_35_pad_0 = const()[name = tensor("current_key_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_35_dilations_0 = const()[name = tensor("current_key_35_dilations_0"), val = tensor([1, 1])]; tensor current_key_35_groups_0 = const()[name = tensor("current_key_35_groups_0"), val = tensor(1)]; tensor layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1029267008)))]; tensor current_key_35_cast_fp16 = conv(dilations = current_key_35_dilations_0, groups = current_key_35_groups_0, pad = current_key_35_pad_0, pad_type = current_key_35_pad_type_0, strides = current_key_35_strides_0, weight = layers_17_self_attn_k_proj_weight_to_fp16, x = obj_239_cast_fp16)[name = tensor("current_key_35_cast_fp16")]; tensor current_value_35_pad_type_0 = const()[name = tensor("current_value_35_pad_type_0"), val = tensor("valid")]; tensor current_value_35_strides_0 = const()[name = tensor("current_value_35_strides_0"), val = tensor([1, 1])]; tensor current_value_35_pad_0 = const()[name = tensor("current_value_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_35_dilations_0 = const()[name = tensor("current_value_35_dilations_0"), val = tensor([1, 1])]; tensor current_value_35_groups_0 = const()[name = tensor("current_value_35_groups_0"), val = tensor(1)]; tensor layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1032543872)))]; tensor layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1035820736)))]; tensor current_value_35_cast_fp16 = conv(bias = layers_17_self_attn_v_proj_bias_to_fp16, dilations = current_value_35_dilations_0, groups = current_value_35_groups_0, pad = current_value_35_pad_0, pad_type = current_value_35_pad_type_0, strides = current_value_35_strides_0, weight = layers_17_self_attn_v_proj_weight_to_fp16, x = obj_239_cast_fp16)[name = tensor("current_value_35_cast_fp16")]; tensor var_3975_cast_fp16 = mul(x = var_103_cast_fp16_17, y = var_239_cast_fp16)[name = tensor("op_3975_cast_fp16")]; tensor var_3976_cast_fp16 = mul(x = current_key_35_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3976_cast_fp16")]; tensor key_69_cast_fp16 = add(x = var_3975_cast_fp16, y = var_3976_cast_fp16)[name = tensor("key_69_cast_fp16")]; tensor var_3979_cast_fp16 = mul(x = var_138_cast_fp16_17, y = var_239_cast_fp16)[name = tensor("op_3979_cast_fp16")]; tensor var_3980_cast_fp16 = mul(x = current_value_35_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_3980_cast_fp16")]; tensor value_69_cast_fp16 = add(x = var_3979_cast_fp16, y = var_3980_cast_fp16)[name = tensor("value_69_cast_fp16")]; tensor var_3984 = const()[name = tensor("op_3984"), val = tensor([1, 20, 64, 1])]; tensor mh_q_69_cast_fp16 = reshape(shape = var_3984, x = query_69_cast_fp16)[name = tensor("mh_q_69_cast_fp16")]; tensor var_3986_to_fp16 = const()[name = tensor("op_3986_to_fp16"), val = tensor(0x1p-3)]; tensor var_3987_cast_fp16 = mul(x = mh_q_69_cast_fp16, y = var_3986_to_fp16)[name = tensor("op_3987_cast_fp16")]; tensor var_3990 = const()[name = tensor("op_3990"), val = tensor([1, 20, 64, 448])]; tensor var_3991_cast_fp16 = reshape(shape = var_3990, x = key_69_cast_fp16)[name = tensor("op_3991_cast_fp16")]; tensor mh_w_103_transpose_x_0 = const()[name = tensor("mh_w_103_transpose_x_0"), val = tensor(true)]; tensor mh_w_103_transpose_y_0 = const()[name = tensor("mh_w_103_transpose_y_0"), val = tensor(false)]; tensor mh_w_103_cast_fp16 = matmul(transpose_x = mh_w_103_transpose_x_0, transpose_y = mh_w_103_transpose_y_0, x = var_3987_cast_fp16, y = var_3991_cast_fp16)[name = tensor("mh_w_103_cast_fp16")]; tensor mh_w_105_cast_fp16 = add(x = mh_w_103_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_105_cast_fp16")]; tensor var_3999_cast_fp16 = softmax(axis = var_3911, x = mh_w_105_cast_fp16)[name = tensor("op_3999_cast_fp16")]; tensor var_4000 = const()[name = tensor("op_4000"), val = tensor([1, 20, 64, 448])]; tensor var_4001_cast_fp16 = reshape(shape = var_4000, x = value_69_cast_fp16)[name = tensor("op_4001_cast_fp16")]; tensor attn_69_transpose_x_0 = const()[name = tensor("attn_69_transpose_x_0"), val = tensor(false)]; tensor attn_69_transpose_y_0 = const()[name = tensor("attn_69_transpose_y_0"), val = tensor(true)]; tensor attn_69_cast_fp16 = matmul(transpose_x = attn_69_transpose_x_0, transpose_y = attn_69_transpose_y_0, x = var_4001_cast_fp16, y = var_3999_cast_fp16)[name = tensor("attn_69_cast_fp16")]; tensor var_4004 = const()[name = tensor("op_4004"), val = tensor([1, 1280, 1, 1])]; tensor input_171_cast_fp16 = reshape(shape = var_4004, x = attn_69_cast_fp16)[name = tensor("input_171_cast_fp16")]; tensor obj_245_pad_type_0 = const()[name = tensor("obj_245_pad_type_0"), val = tensor("valid")]; tensor obj_245_strides_0 = const()[name = tensor("obj_245_strides_0"), val = tensor([1, 1])]; tensor obj_245_pad_0 = const()[name = tensor("obj_245_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_245_dilations_0 = const()[name = tensor("obj_245_dilations_0"), val = tensor([1, 1])]; tensor obj_245_groups_0 = const()[name = tensor("obj_245_groups_0"), val = tensor(1)]; tensor layers_17_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_17_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1035823360)))]; tensor layers_17_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_17_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1039100224)))]; tensor obj_245_cast_fp16 = conv(bias = layers_17_self_attn_o_proj_bias_to_fp16, dilations = obj_245_dilations_0, groups = obj_245_groups_0, pad = obj_245_pad_0, pad_type = obj_245_pad_type_0, strides = obj_245_strides_0, weight = layers_17_self_attn_o_proj_weight_to_fp16, x = input_171_cast_fp16)[name = tensor("obj_245_cast_fp16")]; tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = obj_245_cast_fp16)[name = tensor("inputs_105_cast_fp16")]; tensor out_105_axes_0 = const()[name = tensor("out_105_axes_0"), val = tensor([1])]; tensor var_4026_to_fp16 = const()[name = tensor("op_4026_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_4026_to_fp16, x = inputs_105_cast_fp16)[name = tensor("out_105_cast_fp16")]; tensor obj_247_gamma_0_to_fp16 = const()[name = tensor("obj_247_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1039102848)))]; tensor obj_247_beta_0_to_fp16 = const()[name = tensor("obj_247_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1039105472)))]; tensor obj_247_epsilon_0_to_fp16 = const()[name = tensor("obj_247_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_247_cast_fp16 = batch_norm(beta = obj_247_beta_0_to_fp16, epsilon = obj_247_epsilon_0_to_fp16, gamma = obj_247_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_105_cast_fp16)[name = tensor("obj_247_cast_fp16")]; tensor query_71_pad_type_0 = const()[name = tensor("query_71_pad_type_0"), val = tensor("valid")]; tensor query_71_strides_0 = const()[name = tensor("query_71_strides_0"), val = tensor([1, 1])]; tensor query_71_pad_0 = const()[name = tensor("query_71_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_71_dilations_0 = const()[name = tensor("query_71_dilations_0"), val = tensor([1, 1])]; tensor query_71_groups_0 = const()[name = tensor("query_71_groups_0"), val = tensor(1)]; tensor layers_17_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_17_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1039108096)))]; tensor layers_17_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_17_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1042384960)))]; tensor query_71_cast_fp16 = conv(bias = layers_17_encoder_attn_q_proj_bias_to_fp16, dilations = query_71_dilations_0, groups = query_71_groups_0, pad = query_71_pad_0, pad_type = query_71_pad_type_0, strides = query_71_strides_0, weight = layers_17_encoder_attn_q_proj_weight_to_fp16, x = obj_247_cast_fp16)[name = tensor("query_71_cast_fp16")]; tensor key_71_pad_type_0 = const()[name = tensor("key_71_pad_type_0"), val = tensor("valid")]; tensor key_71_strides_0 = const()[name = tensor("key_71_strides_0"), val = tensor([1, 1])]; tensor key_71_pad_0 = const()[name = tensor("key_71_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_71_dilations_0 = const()[name = tensor("key_71_dilations_0"), val = tensor([1, 1])]; tensor key_71_groups_0 = const()[name = tensor("key_71_groups_0"), val = tensor(1)]; tensor layers_17_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_17_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1042387584)))]; tensor key_71_cast_fp16 = conv(dilations = key_71_dilations_0, groups = key_71_groups_0, pad = key_71_pad_0, pad_type = key_71_pad_type_0, strides = key_71_strides_0, weight = layers_17_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_71_cast_fp16")]; tensor value_71_pad_type_0 = const()[name = tensor("value_71_pad_type_0"), val = tensor("valid")]; tensor value_71_strides_0 = const()[name = tensor("value_71_strides_0"), val = tensor([1, 1])]; tensor value_71_pad_0 = const()[name = tensor("value_71_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_71_dilations_0 = const()[name = tensor("value_71_dilations_0"), val = tensor([1, 1])]; tensor value_71_groups_0 = const()[name = tensor("value_71_groups_0"), val = tensor(1)]; tensor layers_17_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_17_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1045664448)))]; tensor layers_17_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_17_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1048941312)))]; tensor value_71_cast_fp16 = conv(bias = layers_17_encoder_attn_v_proj_bias_to_fp16, dilations = value_71_dilations_0, groups = value_71_groups_0, pad = value_71_pad_0, pad_type = value_71_pad_type_0, strides = value_71_strides_0, weight = layers_17_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_71_cast_fp16")]; tensor var_4062 = const()[name = tensor("op_4062"), val = tensor([1, 20, 64, 1])]; tensor mh_q_71_cast_fp16 = reshape(shape = var_4062, x = query_71_cast_fp16)[name = tensor("mh_q_71_cast_fp16")]; tensor var_4064_to_fp16 = const()[name = tensor("op_4064_to_fp16"), val = tensor(0x1p-3)]; tensor var_4065_cast_fp16 = mul(x = mh_q_71_cast_fp16, y = var_4064_to_fp16)[name = tensor("op_4065_cast_fp16")]; tensor var_4068 = const()[name = tensor("op_4068"), val = tensor([1, 20, 64, 1500])]; tensor var_4069_cast_fp16 = reshape(shape = var_4068, x = key_71_cast_fp16)[name = tensor("op_4069_cast_fp16")]; tensor mh_w_107_transpose_x_0 = const()[name = tensor("mh_w_107_transpose_x_0"), val = tensor(true)]; tensor mh_w_107_transpose_y_0 = const()[name = tensor("mh_w_107_transpose_y_0"), val = tensor(false)]; tensor mh_w_107_cast_fp16 = matmul(transpose_x = mh_w_107_transpose_x_0, transpose_y = mh_w_107_transpose_y_0, x = var_4065_cast_fp16, y = var_4069_cast_fp16)[name = tensor("mh_w_107_cast_fp16")]; tensor obj_251_cast_fp16 = softmax(axis = var_3911, x = mh_w_107_cast_fp16)[name = tensor("obj_251_cast_fp16")]; tensor var_4073 = const()[name = tensor("op_4073"), val = tensor([1, 20, 64, 1500])]; tensor var_4074_cast_fp16 = reshape(shape = var_4073, x = value_71_cast_fp16)[name = tensor("op_4074_cast_fp16")]; tensor attn_71_transpose_x_0 = const()[name = tensor("attn_71_transpose_x_0"), val = tensor(false)]; tensor attn_71_transpose_y_0 = const()[name = tensor("attn_71_transpose_y_0"), val = tensor(true)]; tensor attn_71_cast_fp16 = matmul(transpose_x = attn_71_transpose_x_0, transpose_y = attn_71_transpose_y_0, x = var_4074_cast_fp16, y = obj_251_cast_fp16)[name = tensor("attn_71_cast_fp16")]; tensor var_4077 = const()[name = tensor("op_4077"), val = tensor([1, 1280, 1, 1])]; tensor input_173_cast_fp16 = reshape(shape = var_4077, x = attn_71_cast_fp16)[name = tensor("input_173_cast_fp16")]; tensor obj_249_pad_type_0 = const()[name = tensor("obj_249_pad_type_0"), val = tensor("valid")]; tensor obj_249_strides_0 = const()[name = tensor("obj_249_strides_0"), val = tensor([1, 1])]; tensor obj_249_pad_0 = const()[name = tensor("obj_249_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_249_dilations_0 = const()[name = tensor("obj_249_dilations_0"), val = tensor([1, 1])]; tensor obj_249_groups_0 = const()[name = tensor("obj_249_groups_0"), val = tensor(1)]; tensor layers_17_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_17_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1048943936)))]; tensor layers_17_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_17_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1052220800)))]; tensor obj_249_cast_fp16 = conv(bias = layers_17_encoder_attn_o_proj_bias_to_fp16, dilations = obj_249_dilations_0, groups = obj_249_groups_0, pad = obj_249_pad_0, pad_type = obj_249_pad_type_0, strides = obj_249_strides_0, weight = layers_17_encoder_attn_o_proj_weight_to_fp16, x = input_173_cast_fp16)[name = tensor("obj_249_cast_fp16")]; tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = obj_249_cast_fp16)[name = tensor("inputs_107_cast_fp16")]; tensor out_107_axes_0 = const()[name = tensor("out_107_axes_0"), val = tensor([1])]; tensor var_4098_to_fp16 = const()[name = tensor("op_4098_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_4098_to_fp16, x = inputs_107_cast_fp16)[name = tensor("out_107_cast_fp16")]; tensor input_175_gamma_0_to_fp16 = const()[name = tensor("input_175_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1052223424)))]; tensor input_175_beta_0_to_fp16 = const()[name = tensor("input_175_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1052226048)))]; tensor input_175_epsilon_0_to_fp16 = const()[name = tensor("input_175_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_175_cast_fp16 = batch_norm(beta = input_175_beta_0_to_fp16, epsilon = input_175_epsilon_0_to_fp16, gamma = input_175_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_107_cast_fp16)[name = tensor("input_175_cast_fp16")]; tensor input_177_pad_type_0 = const()[name = tensor("input_177_pad_type_0"), val = tensor("valid")]; tensor input_177_strides_0 = const()[name = tensor("input_177_strides_0"), val = tensor([1, 1])]; tensor input_177_pad_0 = const()[name = tensor("input_177_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_177_dilations_0 = const()[name = tensor("input_177_dilations_0"), val = tensor([1, 1])]; tensor input_177_groups_0 = const()[name = tensor("input_177_groups_0"), val = tensor(1)]; tensor layers_17_fc1_weight_to_fp16 = const()[name = tensor("layers_17_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1052228672)))]; tensor layers_17_fc1_bias_to_fp16 = const()[name = tensor("layers_17_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1065335936)))]; tensor input_177_cast_fp16 = conv(bias = layers_17_fc1_bias_to_fp16, dilations = input_177_dilations_0, groups = input_177_groups_0, pad = input_177_pad_0, pad_type = input_177_pad_type_0, strides = input_177_strides_0, weight = layers_17_fc1_weight_to_fp16, x = input_175_cast_fp16)[name = tensor("input_177_cast_fp16")]; tensor input_179_mode_0 = const()[name = tensor("input_179_mode_0"), val = tensor("EXACT")]; tensor input_179_cast_fp16 = gelu(mode = input_179_mode_0, x = input_177_cast_fp16)[name = tensor("input_179_cast_fp16")]; tensor hidden_states_37_pad_type_0 = const()[name = tensor("hidden_states_37_pad_type_0"), val = tensor("valid")]; tensor hidden_states_37_strides_0 = const()[name = tensor("hidden_states_37_strides_0"), val = tensor([1, 1])]; tensor hidden_states_37_pad_0 = const()[name = tensor("hidden_states_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_37_dilations_0 = const()[name = tensor("hidden_states_37_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_37_groups_0 = const()[name = tensor("hidden_states_37_groups_0"), val = tensor(1)]; tensor layers_17_fc2_weight_to_fp16 = const()[name = tensor("layers_17_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1065346240)))]; tensor layers_17_fc2_bias_to_fp16 = const()[name = tensor("layers_17_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1078453504)))]; tensor hidden_states_37_cast_fp16 = conv(bias = layers_17_fc2_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = layers_17_fc2_weight_to_fp16, x = input_179_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = hidden_states_37_cast_fp16)[name = tensor("inputs_109_cast_fp16")]; tensor var_4134 = const()[name = tensor("op_4134"), val = tensor(3)]; tensor out_109_axes_0 = const()[name = tensor("out_109_axes_0"), val = tensor([1])]; tensor var_4159_to_fp16 = const()[name = tensor("op_4159_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_4159_to_fp16, x = inputs_109_cast_fp16)[name = tensor("out_109_cast_fp16")]; tensor obj_253_gamma_0_to_fp16 = const()[name = tensor("obj_253_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1078456128)))]; tensor obj_253_beta_0_to_fp16 = const()[name = tensor("obj_253_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1078458752)))]; tensor obj_253_epsilon_0_to_fp16 = const()[name = tensor("obj_253_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_253_cast_fp16 = batch_norm(beta = obj_253_beta_0_to_fp16, epsilon = obj_253_epsilon_0_to_fp16, gamma = obj_253_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_109_cast_fp16)[name = tensor("obj_253_cast_fp16")]; tensor query_73_pad_type_0 = const()[name = tensor("query_73_pad_type_0"), val = tensor("valid")]; tensor query_73_strides_0 = const()[name = tensor("query_73_strides_0"), val = tensor([1, 1])]; tensor query_73_pad_0 = const()[name = tensor("query_73_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_73_dilations_0 = const()[name = tensor("query_73_dilations_0"), val = tensor([1, 1])]; tensor query_73_groups_0 = const()[name = tensor("query_73_groups_0"), val = tensor(1)]; tensor layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1078461376)))]; tensor layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1081738240)))]; tensor query_73_cast_fp16 = conv(bias = layers_18_self_attn_q_proj_bias_to_fp16, dilations = query_73_dilations_0, groups = query_73_groups_0, pad = query_73_pad_0, pad_type = query_73_pad_type_0, strides = query_73_strides_0, weight = layers_18_self_attn_q_proj_weight_to_fp16, x = obj_253_cast_fp16)[name = tensor("query_73_cast_fp16")]; tensor current_key_37_pad_type_0 = const()[name = tensor("current_key_37_pad_type_0"), val = tensor("valid")]; tensor current_key_37_strides_0 = const()[name = tensor("current_key_37_strides_0"), val = tensor([1, 1])]; tensor current_key_37_pad_0 = const()[name = tensor("current_key_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_37_dilations_0 = const()[name = tensor("current_key_37_dilations_0"), val = tensor([1, 1])]; tensor current_key_37_groups_0 = const()[name = tensor("current_key_37_groups_0"), val = tensor(1)]; tensor layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1081740864)))]; tensor current_key_37_cast_fp16 = conv(dilations = current_key_37_dilations_0, groups = current_key_37_groups_0, pad = current_key_37_pad_0, pad_type = current_key_37_pad_type_0, strides = current_key_37_strides_0, weight = layers_18_self_attn_k_proj_weight_to_fp16, x = obj_253_cast_fp16)[name = tensor("current_key_37_cast_fp16")]; tensor current_value_37_pad_type_0 = const()[name = tensor("current_value_37_pad_type_0"), val = tensor("valid")]; tensor current_value_37_strides_0 = const()[name = tensor("current_value_37_strides_0"), val = tensor([1, 1])]; tensor current_value_37_pad_0 = const()[name = tensor("current_value_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_37_dilations_0 = const()[name = tensor("current_value_37_dilations_0"), val = tensor([1, 1])]; tensor current_value_37_groups_0 = const()[name = tensor("current_value_37_groups_0"), val = tensor(1)]; tensor layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1085017728)))]; tensor layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1088294592)))]; tensor current_value_37_cast_fp16 = conv(bias = layers_18_self_attn_v_proj_bias_to_fp16, dilations = current_value_37_dilations_0, groups = current_value_37_groups_0, pad = current_value_37_pad_0, pad_type = current_value_37_pad_type_0, strides = current_value_37_strides_0, weight = layers_18_self_attn_v_proj_weight_to_fp16, x = obj_253_cast_fp16)[name = tensor("current_value_37_cast_fp16")]; tensor var_4198_cast_fp16 = mul(x = var_103_cast_fp16_18, y = var_239_cast_fp16)[name = tensor("op_4198_cast_fp16")]; tensor var_4199_cast_fp16 = mul(x = current_key_37_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_4199_cast_fp16")]; tensor key_73_cast_fp16 = add(x = var_4198_cast_fp16, y = var_4199_cast_fp16)[name = tensor("key_73_cast_fp16")]; tensor var_4202_cast_fp16 = mul(x = var_138_cast_fp16_18, y = var_239_cast_fp16)[name = tensor("op_4202_cast_fp16")]; tensor var_4203_cast_fp16 = mul(x = current_value_37_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_4203_cast_fp16")]; tensor value_73_cast_fp16 = add(x = var_4202_cast_fp16, y = var_4203_cast_fp16)[name = tensor("value_73_cast_fp16")]; tensor var_4207 = const()[name = tensor("op_4207"), val = tensor([1, 20, 64, 1])]; tensor mh_q_73_cast_fp16 = reshape(shape = var_4207, x = query_73_cast_fp16)[name = tensor("mh_q_73_cast_fp16")]; tensor var_4209_to_fp16 = const()[name = tensor("op_4209_to_fp16"), val = tensor(0x1p-3)]; tensor var_4210_cast_fp16 = mul(x = mh_q_73_cast_fp16, y = var_4209_to_fp16)[name = tensor("op_4210_cast_fp16")]; tensor var_4213 = const()[name = tensor("op_4213"), val = tensor([1, 20, 64, 448])]; tensor var_4214_cast_fp16 = reshape(shape = var_4213, x = key_73_cast_fp16)[name = tensor("op_4214_cast_fp16")]; tensor mh_w_109_transpose_x_0 = const()[name = tensor("mh_w_109_transpose_x_0"), val = tensor(true)]; tensor mh_w_109_transpose_y_0 = const()[name = tensor("mh_w_109_transpose_y_0"), val = tensor(false)]; tensor mh_w_109_cast_fp16 = matmul(transpose_x = mh_w_109_transpose_x_0, transpose_y = mh_w_109_transpose_y_0, x = var_4210_cast_fp16, y = var_4214_cast_fp16)[name = tensor("mh_w_109_cast_fp16")]; tensor mh_w_111_cast_fp16 = add(x = mh_w_109_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_111_cast_fp16")]; tensor var_4222_cast_fp16 = softmax(axis = var_4134, x = mh_w_111_cast_fp16)[name = tensor("op_4222_cast_fp16")]; tensor var_4223 = const()[name = tensor("op_4223"), val = tensor([1, 20, 64, 448])]; tensor var_4224_cast_fp16 = reshape(shape = var_4223, x = value_73_cast_fp16)[name = tensor("op_4224_cast_fp16")]; tensor attn_73_transpose_x_0 = const()[name = tensor("attn_73_transpose_x_0"), val = tensor(false)]; tensor attn_73_transpose_y_0 = const()[name = tensor("attn_73_transpose_y_0"), val = tensor(true)]; tensor attn_73_cast_fp16 = matmul(transpose_x = attn_73_transpose_x_0, transpose_y = attn_73_transpose_y_0, x = var_4224_cast_fp16, y = var_4222_cast_fp16)[name = tensor("attn_73_cast_fp16")]; tensor var_4227 = const()[name = tensor("op_4227"), val = tensor([1, 1280, 1, 1])]; tensor input_181_cast_fp16 = reshape(shape = var_4227, x = attn_73_cast_fp16)[name = tensor("input_181_cast_fp16")]; tensor obj_259_pad_type_0 = const()[name = tensor("obj_259_pad_type_0"), val = tensor("valid")]; tensor obj_259_strides_0 = const()[name = tensor("obj_259_strides_0"), val = tensor([1, 1])]; tensor obj_259_pad_0 = const()[name = tensor("obj_259_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_259_dilations_0 = const()[name = tensor("obj_259_dilations_0"), val = tensor([1, 1])]; tensor obj_259_groups_0 = const()[name = tensor("obj_259_groups_0"), val = tensor(1)]; tensor layers_18_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_18_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1088297216)))]; tensor layers_18_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_18_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1091574080)))]; tensor obj_259_cast_fp16 = conv(bias = layers_18_self_attn_o_proj_bias_to_fp16, dilations = obj_259_dilations_0, groups = obj_259_groups_0, pad = obj_259_pad_0, pad_type = obj_259_pad_type_0, strides = obj_259_strides_0, weight = layers_18_self_attn_o_proj_weight_to_fp16, x = input_181_cast_fp16)[name = tensor("obj_259_cast_fp16")]; tensor inputs_111_cast_fp16 = add(x = inputs_109_cast_fp16, y = obj_259_cast_fp16)[name = tensor("inputs_111_cast_fp16")]; tensor out_111_axes_0 = const()[name = tensor("out_111_axes_0"), val = tensor([1])]; tensor var_4249_to_fp16 = const()[name = tensor("op_4249_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_4249_to_fp16, x = inputs_111_cast_fp16)[name = tensor("out_111_cast_fp16")]; tensor obj_261_gamma_0_to_fp16 = const()[name = tensor("obj_261_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1091576704)))]; tensor obj_261_beta_0_to_fp16 = const()[name = tensor("obj_261_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1091579328)))]; tensor obj_261_epsilon_0_to_fp16 = const()[name = tensor("obj_261_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_261_cast_fp16 = batch_norm(beta = obj_261_beta_0_to_fp16, epsilon = obj_261_epsilon_0_to_fp16, gamma = obj_261_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_111_cast_fp16)[name = tensor("obj_261_cast_fp16")]; tensor query_75_pad_type_0 = const()[name = tensor("query_75_pad_type_0"), val = tensor("valid")]; tensor query_75_strides_0 = const()[name = tensor("query_75_strides_0"), val = tensor([1, 1])]; tensor query_75_pad_0 = const()[name = tensor("query_75_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_75_dilations_0 = const()[name = tensor("query_75_dilations_0"), val = tensor([1, 1])]; tensor query_75_groups_0 = const()[name = tensor("query_75_groups_0"), val = tensor(1)]; tensor layers_18_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_18_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1091581952)))]; tensor layers_18_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_18_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1094858816)))]; tensor query_75_cast_fp16 = conv(bias = layers_18_encoder_attn_q_proj_bias_to_fp16, dilations = query_75_dilations_0, groups = query_75_groups_0, pad = query_75_pad_0, pad_type = query_75_pad_type_0, strides = query_75_strides_0, weight = layers_18_encoder_attn_q_proj_weight_to_fp16, x = obj_261_cast_fp16)[name = tensor("query_75_cast_fp16")]; tensor key_75_pad_type_0 = const()[name = tensor("key_75_pad_type_0"), val = tensor("valid")]; tensor key_75_strides_0 = const()[name = tensor("key_75_strides_0"), val = tensor([1, 1])]; tensor key_75_pad_0 = const()[name = tensor("key_75_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_75_dilations_0 = const()[name = tensor("key_75_dilations_0"), val = tensor([1, 1])]; tensor key_75_groups_0 = const()[name = tensor("key_75_groups_0"), val = tensor(1)]; tensor layers_18_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_18_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1094861440)))]; tensor key_75_cast_fp16 = conv(dilations = key_75_dilations_0, groups = key_75_groups_0, pad = key_75_pad_0, pad_type = key_75_pad_type_0, strides = key_75_strides_0, weight = layers_18_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_75_cast_fp16")]; tensor value_75_pad_type_0 = const()[name = tensor("value_75_pad_type_0"), val = tensor("valid")]; tensor value_75_strides_0 = const()[name = tensor("value_75_strides_0"), val = tensor([1, 1])]; tensor value_75_pad_0 = const()[name = tensor("value_75_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_75_dilations_0 = const()[name = tensor("value_75_dilations_0"), val = tensor([1, 1])]; tensor value_75_groups_0 = const()[name = tensor("value_75_groups_0"), val = tensor(1)]; tensor layers_18_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_18_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1098138304)))]; tensor layers_18_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_18_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1101415168)))]; tensor value_75_cast_fp16 = conv(bias = layers_18_encoder_attn_v_proj_bias_to_fp16, dilations = value_75_dilations_0, groups = value_75_groups_0, pad = value_75_pad_0, pad_type = value_75_pad_type_0, strides = value_75_strides_0, weight = layers_18_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_75_cast_fp16")]; tensor var_4285 = const()[name = tensor("op_4285"), val = tensor([1, 20, 64, 1])]; tensor mh_q_75_cast_fp16 = reshape(shape = var_4285, x = query_75_cast_fp16)[name = tensor("mh_q_75_cast_fp16")]; tensor var_4287_to_fp16 = const()[name = tensor("op_4287_to_fp16"), val = tensor(0x1p-3)]; tensor var_4288_cast_fp16 = mul(x = mh_q_75_cast_fp16, y = var_4287_to_fp16)[name = tensor("op_4288_cast_fp16")]; tensor var_4291 = const()[name = tensor("op_4291"), val = tensor([1, 20, 64, 1500])]; tensor var_4292_cast_fp16 = reshape(shape = var_4291, x = key_75_cast_fp16)[name = tensor("op_4292_cast_fp16")]; tensor mh_w_113_transpose_x_0 = const()[name = tensor("mh_w_113_transpose_x_0"), val = tensor(true)]; tensor mh_w_113_transpose_y_0 = const()[name = tensor("mh_w_113_transpose_y_0"), val = tensor(false)]; tensor mh_w_113_cast_fp16 = matmul(transpose_x = mh_w_113_transpose_x_0, transpose_y = mh_w_113_transpose_y_0, x = var_4288_cast_fp16, y = var_4292_cast_fp16)[name = tensor("mh_w_113_cast_fp16")]; tensor obj_265_cast_fp16 = softmax(axis = var_4134, x = mh_w_113_cast_fp16)[name = tensor("obj_265_cast_fp16")]; tensor var_4296 = const()[name = tensor("op_4296"), val = tensor([1, 20, 64, 1500])]; tensor var_4297_cast_fp16 = reshape(shape = var_4296, x = value_75_cast_fp16)[name = tensor("op_4297_cast_fp16")]; tensor attn_75_transpose_x_0 = const()[name = tensor("attn_75_transpose_x_0"), val = tensor(false)]; tensor attn_75_transpose_y_0 = const()[name = tensor("attn_75_transpose_y_0"), val = tensor(true)]; tensor attn_75_cast_fp16 = matmul(transpose_x = attn_75_transpose_x_0, transpose_y = attn_75_transpose_y_0, x = var_4297_cast_fp16, y = obj_265_cast_fp16)[name = tensor("attn_75_cast_fp16")]; tensor var_4300 = const()[name = tensor("op_4300"), val = tensor([1, 1280, 1, 1])]; tensor input_183_cast_fp16 = reshape(shape = var_4300, x = attn_75_cast_fp16)[name = tensor("input_183_cast_fp16")]; tensor obj_263_pad_type_0 = const()[name = tensor("obj_263_pad_type_0"), val = tensor("valid")]; tensor obj_263_strides_0 = const()[name = tensor("obj_263_strides_0"), val = tensor([1, 1])]; tensor obj_263_pad_0 = const()[name = tensor("obj_263_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_263_dilations_0 = const()[name = tensor("obj_263_dilations_0"), val = tensor([1, 1])]; tensor obj_263_groups_0 = const()[name = tensor("obj_263_groups_0"), val = tensor(1)]; tensor layers_18_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_18_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1101417792)))]; tensor layers_18_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_18_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1104694656)))]; tensor obj_263_cast_fp16 = conv(bias = layers_18_encoder_attn_o_proj_bias_to_fp16, dilations = obj_263_dilations_0, groups = obj_263_groups_0, pad = obj_263_pad_0, pad_type = obj_263_pad_type_0, strides = obj_263_strides_0, weight = layers_18_encoder_attn_o_proj_weight_to_fp16, x = input_183_cast_fp16)[name = tensor("obj_263_cast_fp16")]; tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = obj_263_cast_fp16)[name = tensor("inputs_113_cast_fp16")]; tensor out_113_axes_0 = const()[name = tensor("out_113_axes_0"), val = tensor([1])]; tensor var_4321_to_fp16 = const()[name = tensor("op_4321_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_4321_to_fp16, x = inputs_113_cast_fp16)[name = tensor("out_113_cast_fp16")]; tensor input_185_gamma_0_to_fp16 = const()[name = tensor("input_185_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1104697280)))]; tensor input_185_beta_0_to_fp16 = const()[name = tensor("input_185_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1104699904)))]; tensor input_185_epsilon_0_to_fp16 = const()[name = tensor("input_185_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_185_cast_fp16 = batch_norm(beta = input_185_beta_0_to_fp16, epsilon = input_185_epsilon_0_to_fp16, gamma = input_185_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_113_cast_fp16)[name = tensor("input_185_cast_fp16")]; tensor input_187_pad_type_0 = const()[name = tensor("input_187_pad_type_0"), val = tensor("valid")]; tensor input_187_strides_0 = const()[name = tensor("input_187_strides_0"), val = tensor([1, 1])]; tensor input_187_pad_0 = const()[name = tensor("input_187_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_187_dilations_0 = const()[name = tensor("input_187_dilations_0"), val = tensor([1, 1])]; tensor input_187_groups_0 = const()[name = tensor("input_187_groups_0"), val = tensor(1)]; tensor layers_18_fc1_weight_to_fp16 = const()[name = tensor("layers_18_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1104702528)))]; tensor layers_18_fc1_bias_to_fp16 = const()[name = tensor("layers_18_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1117809792)))]; tensor input_187_cast_fp16 = conv(bias = layers_18_fc1_bias_to_fp16, dilations = input_187_dilations_0, groups = input_187_groups_0, pad = input_187_pad_0, pad_type = input_187_pad_type_0, strides = input_187_strides_0, weight = layers_18_fc1_weight_to_fp16, x = input_185_cast_fp16)[name = tensor("input_187_cast_fp16")]; tensor input_189_mode_0 = const()[name = tensor("input_189_mode_0"), val = tensor("EXACT")]; tensor input_189_cast_fp16 = gelu(mode = input_189_mode_0, x = input_187_cast_fp16)[name = tensor("input_189_cast_fp16")]; tensor hidden_states_39_pad_type_0 = const()[name = tensor("hidden_states_39_pad_type_0"), val = tensor("valid")]; tensor hidden_states_39_strides_0 = const()[name = tensor("hidden_states_39_strides_0"), val = tensor([1, 1])]; tensor hidden_states_39_pad_0 = const()[name = tensor("hidden_states_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_39_dilations_0 = const()[name = tensor("hidden_states_39_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_39_groups_0 = const()[name = tensor("hidden_states_39_groups_0"), val = tensor(1)]; tensor layers_18_fc2_weight_to_fp16 = const()[name = tensor("layers_18_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1117820096)))]; tensor layers_18_fc2_bias_to_fp16 = const()[name = tensor("layers_18_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1130927360)))]; tensor hidden_states_39_cast_fp16 = conv(bias = layers_18_fc2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = layers_18_fc2_weight_to_fp16, x = input_189_cast_fp16)[name = tensor("hidden_states_39_cast_fp16")]; tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = hidden_states_39_cast_fp16)[name = tensor("inputs_115_cast_fp16")]; tensor var_4357 = const()[name = tensor("op_4357"), val = tensor(3)]; tensor out_115_axes_0 = const()[name = tensor("out_115_axes_0"), val = tensor([1])]; tensor var_4382_to_fp16 = const()[name = tensor("op_4382_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_4382_to_fp16, x = inputs_115_cast_fp16)[name = tensor("out_115_cast_fp16")]; tensor obj_267_gamma_0_to_fp16 = const()[name = tensor("obj_267_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1130929984)))]; tensor obj_267_beta_0_to_fp16 = const()[name = tensor("obj_267_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1130932608)))]; tensor obj_267_epsilon_0_to_fp16 = const()[name = tensor("obj_267_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_267_cast_fp16 = batch_norm(beta = obj_267_beta_0_to_fp16, epsilon = obj_267_epsilon_0_to_fp16, gamma = obj_267_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_115_cast_fp16)[name = tensor("obj_267_cast_fp16")]; tensor query_77_pad_type_0 = const()[name = tensor("query_77_pad_type_0"), val = tensor("valid")]; tensor query_77_strides_0 = const()[name = tensor("query_77_strides_0"), val = tensor([1, 1])]; tensor query_77_pad_0 = const()[name = tensor("query_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_77_dilations_0 = const()[name = tensor("query_77_dilations_0"), val = tensor([1, 1])]; tensor query_77_groups_0 = const()[name = tensor("query_77_groups_0"), val = tensor(1)]; tensor layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1130935232)))]; tensor layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1134212096)))]; tensor query_77_cast_fp16 = conv(bias = layers_19_self_attn_q_proj_bias_to_fp16, dilations = query_77_dilations_0, groups = query_77_groups_0, pad = query_77_pad_0, pad_type = query_77_pad_type_0, strides = query_77_strides_0, weight = layers_19_self_attn_q_proj_weight_to_fp16, x = obj_267_cast_fp16)[name = tensor("query_77_cast_fp16")]; tensor current_key_39_pad_type_0 = const()[name = tensor("current_key_39_pad_type_0"), val = tensor("valid")]; tensor current_key_39_strides_0 = const()[name = tensor("current_key_39_strides_0"), val = tensor([1, 1])]; tensor current_key_39_pad_0 = const()[name = tensor("current_key_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_39_dilations_0 = const()[name = tensor("current_key_39_dilations_0"), val = tensor([1, 1])]; tensor current_key_39_groups_0 = const()[name = tensor("current_key_39_groups_0"), val = tensor(1)]; tensor layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1134214720)))]; tensor current_key_39_cast_fp16 = conv(dilations = current_key_39_dilations_0, groups = current_key_39_groups_0, pad = current_key_39_pad_0, pad_type = current_key_39_pad_type_0, strides = current_key_39_strides_0, weight = layers_19_self_attn_k_proj_weight_to_fp16, x = obj_267_cast_fp16)[name = tensor("current_key_39_cast_fp16")]; tensor current_value_39_pad_type_0 = const()[name = tensor("current_value_39_pad_type_0"), val = tensor("valid")]; tensor current_value_39_strides_0 = const()[name = tensor("current_value_39_strides_0"), val = tensor([1, 1])]; tensor current_value_39_pad_0 = const()[name = tensor("current_value_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_39_dilations_0 = const()[name = tensor("current_value_39_dilations_0"), val = tensor([1, 1])]; tensor current_value_39_groups_0 = const()[name = tensor("current_value_39_groups_0"), val = tensor(1)]; tensor layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1137491584)))]; tensor layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1140768448)))]; tensor current_value_39_cast_fp16 = conv(bias = layers_19_self_attn_v_proj_bias_to_fp16, dilations = current_value_39_dilations_0, groups = current_value_39_groups_0, pad = current_value_39_pad_0, pad_type = current_value_39_pad_type_0, strides = current_value_39_strides_0, weight = layers_19_self_attn_v_proj_weight_to_fp16, x = obj_267_cast_fp16)[name = tensor("current_value_39_cast_fp16")]; tensor var_4421_cast_fp16 = mul(x = var_103_cast_fp16_19, y = var_239_cast_fp16)[name = tensor("op_4421_cast_fp16")]; tensor var_4422_cast_fp16 = mul(x = current_key_39_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_4422_cast_fp16")]; tensor key_77_cast_fp16 = add(x = var_4421_cast_fp16, y = var_4422_cast_fp16)[name = tensor("key_77_cast_fp16")]; tensor var_4425_cast_fp16 = mul(x = var_138_cast_fp16_19, y = var_239_cast_fp16)[name = tensor("op_4425_cast_fp16")]; tensor var_4426_cast_fp16 = mul(x = current_value_39_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_4426_cast_fp16")]; tensor value_77_cast_fp16 = add(x = var_4425_cast_fp16, y = var_4426_cast_fp16)[name = tensor("value_77_cast_fp16")]; tensor var_4430 = const()[name = tensor("op_4430"), val = tensor([1, 20, 64, 1])]; tensor mh_q_77_cast_fp16 = reshape(shape = var_4430, x = query_77_cast_fp16)[name = tensor("mh_q_77_cast_fp16")]; tensor var_4432_to_fp16 = const()[name = tensor("op_4432_to_fp16"), val = tensor(0x1p-3)]; tensor var_4433_cast_fp16 = mul(x = mh_q_77_cast_fp16, y = var_4432_to_fp16)[name = tensor("op_4433_cast_fp16")]; tensor var_4436 = const()[name = tensor("op_4436"), val = tensor([1, 20, 64, 448])]; tensor var_4437_cast_fp16 = reshape(shape = var_4436, x = key_77_cast_fp16)[name = tensor("op_4437_cast_fp16")]; tensor mh_w_115_transpose_x_0 = const()[name = tensor("mh_w_115_transpose_x_0"), val = tensor(true)]; tensor mh_w_115_transpose_y_0 = const()[name = tensor("mh_w_115_transpose_y_0"), val = tensor(false)]; tensor mh_w_115_cast_fp16 = matmul(transpose_x = mh_w_115_transpose_x_0, transpose_y = mh_w_115_transpose_y_0, x = var_4433_cast_fp16, y = var_4437_cast_fp16)[name = tensor("mh_w_115_cast_fp16")]; tensor mh_w_117_cast_fp16 = add(x = mh_w_115_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_117_cast_fp16")]; tensor var_4445_cast_fp16 = softmax(axis = var_4357, x = mh_w_117_cast_fp16)[name = tensor("op_4445_cast_fp16")]; tensor var_4446 = const()[name = tensor("op_4446"), val = tensor([1, 20, 64, 448])]; tensor var_4447_cast_fp16 = reshape(shape = var_4446, x = value_77_cast_fp16)[name = tensor("op_4447_cast_fp16")]; tensor attn_77_transpose_x_0 = const()[name = tensor("attn_77_transpose_x_0"), val = tensor(false)]; tensor attn_77_transpose_y_0 = const()[name = tensor("attn_77_transpose_y_0"), val = tensor(true)]; tensor attn_77_cast_fp16 = matmul(transpose_x = attn_77_transpose_x_0, transpose_y = attn_77_transpose_y_0, x = var_4447_cast_fp16, y = var_4445_cast_fp16)[name = tensor("attn_77_cast_fp16")]; tensor var_4450 = const()[name = tensor("op_4450"), val = tensor([1, 1280, 1, 1])]; tensor input_191_cast_fp16 = reshape(shape = var_4450, x = attn_77_cast_fp16)[name = tensor("input_191_cast_fp16")]; tensor obj_273_pad_type_0 = const()[name = tensor("obj_273_pad_type_0"), val = tensor("valid")]; tensor obj_273_strides_0 = const()[name = tensor("obj_273_strides_0"), val = tensor([1, 1])]; tensor obj_273_pad_0 = const()[name = tensor("obj_273_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_273_dilations_0 = const()[name = tensor("obj_273_dilations_0"), val = tensor([1, 1])]; tensor obj_273_groups_0 = const()[name = tensor("obj_273_groups_0"), val = tensor(1)]; tensor layers_19_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_19_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1140771072)))]; tensor layers_19_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_19_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1144047936)))]; tensor obj_273_cast_fp16 = conv(bias = layers_19_self_attn_o_proj_bias_to_fp16, dilations = obj_273_dilations_0, groups = obj_273_groups_0, pad = obj_273_pad_0, pad_type = obj_273_pad_type_0, strides = obj_273_strides_0, weight = layers_19_self_attn_o_proj_weight_to_fp16, x = input_191_cast_fp16)[name = tensor("obj_273_cast_fp16")]; tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = obj_273_cast_fp16)[name = tensor("inputs_117_cast_fp16")]; tensor out_117_axes_0 = const()[name = tensor("out_117_axes_0"), val = tensor([1])]; tensor var_4472_to_fp16 = const()[name = tensor("op_4472_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_4472_to_fp16, x = inputs_117_cast_fp16)[name = tensor("out_117_cast_fp16")]; tensor obj_275_gamma_0_to_fp16 = const()[name = tensor("obj_275_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1144050560)))]; tensor obj_275_beta_0_to_fp16 = const()[name = tensor("obj_275_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1144053184)))]; tensor obj_275_epsilon_0_to_fp16 = const()[name = tensor("obj_275_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_275_cast_fp16 = batch_norm(beta = obj_275_beta_0_to_fp16, epsilon = obj_275_epsilon_0_to_fp16, gamma = obj_275_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_117_cast_fp16)[name = tensor("obj_275_cast_fp16")]; tensor query_79_pad_type_0 = const()[name = tensor("query_79_pad_type_0"), val = tensor("valid")]; tensor query_79_strides_0 = const()[name = tensor("query_79_strides_0"), val = tensor([1, 1])]; tensor query_79_pad_0 = const()[name = tensor("query_79_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_79_dilations_0 = const()[name = tensor("query_79_dilations_0"), val = tensor([1, 1])]; tensor query_79_groups_0 = const()[name = tensor("query_79_groups_0"), val = tensor(1)]; tensor layers_19_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_19_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1144055808)))]; tensor layers_19_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_19_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1147332672)))]; tensor query_79_cast_fp16 = conv(bias = layers_19_encoder_attn_q_proj_bias_to_fp16, dilations = query_79_dilations_0, groups = query_79_groups_0, pad = query_79_pad_0, pad_type = query_79_pad_type_0, strides = query_79_strides_0, weight = layers_19_encoder_attn_q_proj_weight_to_fp16, x = obj_275_cast_fp16)[name = tensor("query_79_cast_fp16")]; tensor key_79_pad_type_0 = const()[name = tensor("key_79_pad_type_0"), val = tensor("valid")]; tensor key_79_strides_0 = const()[name = tensor("key_79_strides_0"), val = tensor([1, 1])]; tensor key_79_pad_0 = const()[name = tensor("key_79_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_79_dilations_0 = const()[name = tensor("key_79_dilations_0"), val = tensor([1, 1])]; tensor key_79_groups_0 = const()[name = tensor("key_79_groups_0"), val = tensor(1)]; tensor layers_19_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_19_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1147335296)))]; tensor key_79_cast_fp16 = conv(dilations = key_79_dilations_0, groups = key_79_groups_0, pad = key_79_pad_0, pad_type = key_79_pad_type_0, strides = key_79_strides_0, weight = layers_19_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_79_cast_fp16")]; tensor value_79_pad_type_0 = const()[name = tensor("value_79_pad_type_0"), val = tensor("valid")]; tensor value_79_strides_0 = const()[name = tensor("value_79_strides_0"), val = tensor([1, 1])]; tensor value_79_pad_0 = const()[name = tensor("value_79_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_79_dilations_0 = const()[name = tensor("value_79_dilations_0"), val = tensor([1, 1])]; tensor value_79_groups_0 = const()[name = tensor("value_79_groups_0"), val = tensor(1)]; tensor layers_19_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_19_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1150612160)))]; tensor layers_19_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_19_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1153889024)))]; tensor value_79_cast_fp16 = conv(bias = layers_19_encoder_attn_v_proj_bias_to_fp16, dilations = value_79_dilations_0, groups = value_79_groups_0, pad = value_79_pad_0, pad_type = value_79_pad_type_0, strides = value_79_strides_0, weight = layers_19_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_79_cast_fp16")]; tensor var_4508 = const()[name = tensor("op_4508"), val = tensor([1, 20, 64, 1])]; tensor mh_q_79_cast_fp16 = reshape(shape = var_4508, x = query_79_cast_fp16)[name = tensor("mh_q_79_cast_fp16")]; tensor var_4510_to_fp16 = const()[name = tensor("op_4510_to_fp16"), val = tensor(0x1p-3)]; tensor var_4511_cast_fp16 = mul(x = mh_q_79_cast_fp16, y = var_4510_to_fp16)[name = tensor("op_4511_cast_fp16")]; tensor var_4514 = const()[name = tensor("op_4514"), val = tensor([1, 20, 64, 1500])]; tensor var_4515_cast_fp16 = reshape(shape = var_4514, x = key_79_cast_fp16)[name = tensor("op_4515_cast_fp16")]; tensor mh_w_119_transpose_x_0 = const()[name = tensor("mh_w_119_transpose_x_0"), val = tensor(true)]; tensor mh_w_119_transpose_y_0 = const()[name = tensor("mh_w_119_transpose_y_0"), val = tensor(false)]; tensor mh_w_119_cast_fp16 = matmul(transpose_x = mh_w_119_transpose_x_0, transpose_y = mh_w_119_transpose_y_0, x = var_4511_cast_fp16, y = var_4515_cast_fp16)[name = tensor("mh_w_119_cast_fp16")]; tensor obj_279_cast_fp16 = softmax(axis = var_4357, x = mh_w_119_cast_fp16)[name = tensor("obj_279_cast_fp16")]; tensor var_4519 = const()[name = tensor("op_4519"), val = tensor([1, 20, 64, 1500])]; tensor var_4520_cast_fp16 = reshape(shape = var_4519, x = value_79_cast_fp16)[name = tensor("op_4520_cast_fp16")]; tensor attn_79_transpose_x_0 = const()[name = tensor("attn_79_transpose_x_0"), val = tensor(false)]; tensor attn_79_transpose_y_0 = const()[name = tensor("attn_79_transpose_y_0"), val = tensor(true)]; tensor attn_79_cast_fp16 = matmul(transpose_x = attn_79_transpose_x_0, transpose_y = attn_79_transpose_y_0, x = var_4520_cast_fp16, y = obj_279_cast_fp16)[name = tensor("attn_79_cast_fp16")]; tensor var_4523 = const()[name = tensor("op_4523"), val = tensor([1, 1280, 1, 1])]; tensor input_193_cast_fp16 = reshape(shape = var_4523, x = attn_79_cast_fp16)[name = tensor("input_193_cast_fp16")]; tensor obj_277_pad_type_0 = const()[name = tensor("obj_277_pad_type_0"), val = tensor("valid")]; tensor obj_277_strides_0 = const()[name = tensor("obj_277_strides_0"), val = tensor([1, 1])]; tensor obj_277_pad_0 = const()[name = tensor("obj_277_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_277_dilations_0 = const()[name = tensor("obj_277_dilations_0"), val = tensor([1, 1])]; tensor obj_277_groups_0 = const()[name = tensor("obj_277_groups_0"), val = tensor(1)]; tensor layers_19_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_19_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1153891648)))]; tensor layers_19_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_19_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1157168512)))]; tensor obj_277_cast_fp16 = conv(bias = layers_19_encoder_attn_o_proj_bias_to_fp16, dilations = obj_277_dilations_0, groups = obj_277_groups_0, pad = obj_277_pad_0, pad_type = obj_277_pad_type_0, strides = obj_277_strides_0, weight = layers_19_encoder_attn_o_proj_weight_to_fp16, x = input_193_cast_fp16)[name = tensor("obj_277_cast_fp16")]; tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = obj_277_cast_fp16)[name = tensor("inputs_119_cast_fp16")]; tensor out_119_axes_0 = const()[name = tensor("out_119_axes_0"), val = tensor([1])]; tensor var_4544_to_fp16 = const()[name = tensor("op_4544_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_4544_to_fp16, x = inputs_119_cast_fp16)[name = tensor("out_119_cast_fp16")]; tensor input_195_gamma_0_to_fp16 = const()[name = tensor("input_195_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1157171136)))]; tensor input_195_beta_0_to_fp16 = const()[name = tensor("input_195_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1157173760)))]; tensor input_195_epsilon_0_to_fp16 = const()[name = tensor("input_195_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_195_cast_fp16 = batch_norm(beta = input_195_beta_0_to_fp16, epsilon = input_195_epsilon_0_to_fp16, gamma = input_195_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_119_cast_fp16)[name = tensor("input_195_cast_fp16")]; tensor input_197_pad_type_0 = const()[name = tensor("input_197_pad_type_0"), val = tensor("valid")]; tensor input_197_strides_0 = const()[name = tensor("input_197_strides_0"), val = tensor([1, 1])]; tensor input_197_pad_0 = const()[name = tensor("input_197_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_197_dilations_0 = const()[name = tensor("input_197_dilations_0"), val = tensor([1, 1])]; tensor input_197_groups_0 = const()[name = tensor("input_197_groups_0"), val = tensor(1)]; tensor layers_19_fc1_weight_to_fp16 = const()[name = tensor("layers_19_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1157176384)))]; tensor layers_19_fc1_bias_to_fp16 = const()[name = tensor("layers_19_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1170283648)))]; tensor input_197_cast_fp16 = conv(bias = layers_19_fc1_bias_to_fp16, dilations = input_197_dilations_0, groups = input_197_groups_0, pad = input_197_pad_0, pad_type = input_197_pad_type_0, strides = input_197_strides_0, weight = layers_19_fc1_weight_to_fp16, x = input_195_cast_fp16)[name = tensor("input_197_cast_fp16")]; tensor input_199_mode_0 = const()[name = tensor("input_199_mode_0"), val = tensor("EXACT")]; tensor input_199_cast_fp16 = gelu(mode = input_199_mode_0, x = input_197_cast_fp16)[name = tensor("input_199_cast_fp16")]; tensor hidden_states_41_pad_type_0 = const()[name = tensor("hidden_states_41_pad_type_0"), val = tensor("valid")]; tensor hidden_states_41_strides_0 = const()[name = tensor("hidden_states_41_strides_0"), val = tensor([1, 1])]; tensor hidden_states_41_pad_0 = const()[name = tensor("hidden_states_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_41_dilations_0 = const()[name = tensor("hidden_states_41_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_41_groups_0 = const()[name = tensor("hidden_states_41_groups_0"), val = tensor(1)]; tensor layers_19_fc2_weight_to_fp16 = const()[name = tensor("layers_19_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1170293952)))]; tensor layers_19_fc2_bias_to_fp16 = const()[name = tensor("layers_19_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1183401216)))]; tensor hidden_states_41_cast_fp16 = conv(bias = layers_19_fc2_bias_to_fp16, dilations = hidden_states_41_dilations_0, groups = hidden_states_41_groups_0, pad = hidden_states_41_pad_0, pad_type = hidden_states_41_pad_type_0, strides = hidden_states_41_strides_0, weight = layers_19_fc2_weight_to_fp16, x = input_199_cast_fp16)[name = tensor("hidden_states_41_cast_fp16")]; tensor inputs_121_cast_fp16 = add(x = inputs_119_cast_fp16, y = hidden_states_41_cast_fp16)[name = tensor("inputs_121_cast_fp16")]; tensor var_4580 = const()[name = tensor("op_4580"), val = tensor(3)]; tensor out_121_axes_0 = const()[name = tensor("out_121_axes_0"), val = tensor([1])]; tensor var_4605_to_fp16 = const()[name = tensor("op_4605_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_4605_to_fp16, x = inputs_121_cast_fp16)[name = tensor("out_121_cast_fp16")]; tensor obj_281_gamma_0_to_fp16 = const()[name = tensor("obj_281_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1183403840)))]; tensor obj_281_beta_0_to_fp16 = const()[name = tensor("obj_281_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1183406464)))]; tensor obj_281_epsilon_0_to_fp16 = const()[name = tensor("obj_281_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_281_cast_fp16 = batch_norm(beta = obj_281_beta_0_to_fp16, epsilon = obj_281_epsilon_0_to_fp16, gamma = obj_281_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_121_cast_fp16)[name = tensor("obj_281_cast_fp16")]; tensor query_81_pad_type_0 = const()[name = tensor("query_81_pad_type_0"), val = tensor("valid")]; tensor query_81_strides_0 = const()[name = tensor("query_81_strides_0"), val = tensor([1, 1])]; tensor query_81_pad_0 = const()[name = tensor("query_81_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_81_dilations_0 = const()[name = tensor("query_81_dilations_0"), val = tensor([1, 1])]; tensor query_81_groups_0 = const()[name = tensor("query_81_groups_0"), val = tensor(1)]; tensor layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1183409088)))]; tensor layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1186685952)))]; tensor query_81_cast_fp16 = conv(bias = layers_20_self_attn_q_proj_bias_to_fp16, dilations = query_81_dilations_0, groups = query_81_groups_0, pad = query_81_pad_0, pad_type = query_81_pad_type_0, strides = query_81_strides_0, weight = layers_20_self_attn_q_proj_weight_to_fp16, x = obj_281_cast_fp16)[name = tensor("query_81_cast_fp16")]; tensor current_key_41_pad_type_0 = const()[name = tensor("current_key_41_pad_type_0"), val = tensor("valid")]; tensor current_key_41_strides_0 = const()[name = tensor("current_key_41_strides_0"), val = tensor([1, 1])]; tensor current_key_41_pad_0 = const()[name = tensor("current_key_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_41_dilations_0 = const()[name = tensor("current_key_41_dilations_0"), val = tensor([1, 1])]; tensor current_key_41_groups_0 = const()[name = tensor("current_key_41_groups_0"), val = tensor(1)]; tensor layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1186688576)))]; tensor current_key_41_cast_fp16 = conv(dilations = current_key_41_dilations_0, groups = current_key_41_groups_0, pad = current_key_41_pad_0, pad_type = current_key_41_pad_type_0, strides = current_key_41_strides_0, weight = layers_20_self_attn_k_proj_weight_to_fp16, x = obj_281_cast_fp16)[name = tensor("current_key_41_cast_fp16")]; tensor current_value_41_pad_type_0 = const()[name = tensor("current_value_41_pad_type_0"), val = tensor("valid")]; tensor current_value_41_strides_0 = const()[name = tensor("current_value_41_strides_0"), val = tensor([1, 1])]; tensor current_value_41_pad_0 = const()[name = tensor("current_value_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_41_dilations_0 = const()[name = tensor("current_value_41_dilations_0"), val = tensor([1, 1])]; tensor current_value_41_groups_0 = const()[name = tensor("current_value_41_groups_0"), val = tensor(1)]; tensor layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1189965440)))]; tensor layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1193242304)))]; tensor current_value_41_cast_fp16 = conv(bias = layers_20_self_attn_v_proj_bias_to_fp16, dilations = current_value_41_dilations_0, groups = current_value_41_groups_0, pad = current_value_41_pad_0, pad_type = current_value_41_pad_type_0, strides = current_value_41_strides_0, weight = layers_20_self_attn_v_proj_weight_to_fp16, x = obj_281_cast_fp16)[name = tensor("current_value_41_cast_fp16")]; tensor var_4644_cast_fp16 = mul(x = var_103_cast_fp16_20, y = var_239_cast_fp16)[name = tensor("op_4644_cast_fp16")]; tensor var_4645_cast_fp16 = mul(x = current_key_41_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_4645_cast_fp16")]; tensor key_81_cast_fp16 = add(x = var_4644_cast_fp16, y = var_4645_cast_fp16)[name = tensor("key_81_cast_fp16")]; tensor var_4648_cast_fp16 = mul(x = var_138_cast_fp16_20, y = var_239_cast_fp16)[name = tensor("op_4648_cast_fp16")]; tensor var_4649_cast_fp16 = mul(x = current_value_41_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_4649_cast_fp16")]; tensor value_81_cast_fp16 = add(x = var_4648_cast_fp16, y = var_4649_cast_fp16)[name = tensor("value_81_cast_fp16")]; tensor var_4653 = const()[name = tensor("op_4653"), val = tensor([1, 20, 64, 1])]; tensor mh_q_81_cast_fp16 = reshape(shape = var_4653, x = query_81_cast_fp16)[name = tensor("mh_q_81_cast_fp16")]; tensor var_4655_to_fp16 = const()[name = tensor("op_4655_to_fp16"), val = tensor(0x1p-3)]; tensor var_4656_cast_fp16 = mul(x = mh_q_81_cast_fp16, y = var_4655_to_fp16)[name = tensor("op_4656_cast_fp16")]; tensor var_4659 = const()[name = tensor("op_4659"), val = tensor([1, 20, 64, 448])]; tensor var_4660_cast_fp16 = reshape(shape = var_4659, x = key_81_cast_fp16)[name = tensor("op_4660_cast_fp16")]; tensor mh_w_121_transpose_x_0 = const()[name = tensor("mh_w_121_transpose_x_0"), val = tensor(true)]; tensor mh_w_121_transpose_y_0 = const()[name = tensor("mh_w_121_transpose_y_0"), val = tensor(false)]; tensor mh_w_121_cast_fp16 = matmul(transpose_x = mh_w_121_transpose_x_0, transpose_y = mh_w_121_transpose_y_0, x = var_4656_cast_fp16, y = var_4660_cast_fp16)[name = tensor("mh_w_121_cast_fp16")]; tensor mh_w_123_cast_fp16 = add(x = mh_w_121_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_123_cast_fp16")]; tensor var_4668_cast_fp16 = softmax(axis = var_4580, x = mh_w_123_cast_fp16)[name = tensor("op_4668_cast_fp16")]; tensor var_4669 = const()[name = tensor("op_4669"), val = tensor([1, 20, 64, 448])]; tensor var_4670_cast_fp16 = reshape(shape = var_4669, x = value_81_cast_fp16)[name = tensor("op_4670_cast_fp16")]; tensor attn_81_transpose_x_0 = const()[name = tensor("attn_81_transpose_x_0"), val = tensor(false)]; tensor attn_81_transpose_y_0 = const()[name = tensor("attn_81_transpose_y_0"), val = tensor(true)]; tensor attn_81_cast_fp16 = matmul(transpose_x = attn_81_transpose_x_0, transpose_y = attn_81_transpose_y_0, x = var_4670_cast_fp16, y = var_4668_cast_fp16)[name = tensor("attn_81_cast_fp16")]; tensor var_4673 = const()[name = tensor("op_4673"), val = tensor([1, 1280, 1, 1])]; tensor input_201_cast_fp16 = reshape(shape = var_4673, x = attn_81_cast_fp16)[name = tensor("input_201_cast_fp16")]; tensor obj_287_pad_type_0 = const()[name = tensor("obj_287_pad_type_0"), val = tensor("valid")]; tensor obj_287_strides_0 = const()[name = tensor("obj_287_strides_0"), val = tensor([1, 1])]; tensor obj_287_pad_0 = const()[name = tensor("obj_287_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_287_dilations_0 = const()[name = tensor("obj_287_dilations_0"), val = tensor([1, 1])]; tensor obj_287_groups_0 = const()[name = tensor("obj_287_groups_0"), val = tensor(1)]; tensor layers_20_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_20_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1193244928)))]; tensor layers_20_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_20_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1196521792)))]; tensor obj_287_cast_fp16 = conv(bias = layers_20_self_attn_o_proj_bias_to_fp16, dilations = obj_287_dilations_0, groups = obj_287_groups_0, pad = obj_287_pad_0, pad_type = obj_287_pad_type_0, strides = obj_287_strides_0, weight = layers_20_self_attn_o_proj_weight_to_fp16, x = input_201_cast_fp16)[name = tensor("obj_287_cast_fp16")]; tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = obj_287_cast_fp16)[name = tensor("inputs_123_cast_fp16")]; tensor out_123_axes_0 = const()[name = tensor("out_123_axes_0"), val = tensor([1])]; tensor var_4695_to_fp16 = const()[name = tensor("op_4695_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_4695_to_fp16, x = inputs_123_cast_fp16)[name = tensor("out_123_cast_fp16")]; tensor obj_289_gamma_0_to_fp16 = const()[name = tensor("obj_289_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1196524416)))]; tensor obj_289_beta_0_to_fp16 = const()[name = tensor("obj_289_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1196527040)))]; tensor obj_289_epsilon_0_to_fp16 = const()[name = tensor("obj_289_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_289_cast_fp16 = batch_norm(beta = obj_289_beta_0_to_fp16, epsilon = obj_289_epsilon_0_to_fp16, gamma = obj_289_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_123_cast_fp16)[name = tensor("obj_289_cast_fp16")]; tensor query_83_pad_type_0 = const()[name = tensor("query_83_pad_type_0"), val = tensor("valid")]; tensor query_83_strides_0 = const()[name = tensor("query_83_strides_0"), val = tensor([1, 1])]; tensor query_83_pad_0 = const()[name = tensor("query_83_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_83_dilations_0 = const()[name = tensor("query_83_dilations_0"), val = tensor([1, 1])]; tensor query_83_groups_0 = const()[name = tensor("query_83_groups_0"), val = tensor(1)]; tensor layers_20_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_20_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1196529664)))]; tensor layers_20_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_20_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1199806528)))]; tensor query_83_cast_fp16 = conv(bias = layers_20_encoder_attn_q_proj_bias_to_fp16, dilations = query_83_dilations_0, groups = query_83_groups_0, pad = query_83_pad_0, pad_type = query_83_pad_type_0, strides = query_83_strides_0, weight = layers_20_encoder_attn_q_proj_weight_to_fp16, x = obj_289_cast_fp16)[name = tensor("query_83_cast_fp16")]; tensor key_83_pad_type_0 = const()[name = tensor("key_83_pad_type_0"), val = tensor("valid")]; tensor key_83_strides_0 = const()[name = tensor("key_83_strides_0"), val = tensor([1, 1])]; tensor key_83_pad_0 = const()[name = tensor("key_83_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_83_dilations_0 = const()[name = tensor("key_83_dilations_0"), val = tensor([1, 1])]; tensor key_83_groups_0 = const()[name = tensor("key_83_groups_0"), val = tensor(1)]; tensor layers_20_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_20_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1199809152)))]; tensor key_83_cast_fp16 = conv(dilations = key_83_dilations_0, groups = key_83_groups_0, pad = key_83_pad_0, pad_type = key_83_pad_type_0, strides = key_83_strides_0, weight = layers_20_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_83_cast_fp16")]; tensor value_83_pad_type_0 = const()[name = tensor("value_83_pad_type_0"), val = tensor("valid")]; tensor value_83_strides_0 = const()[name = tensor("value_83_strides_0"), val = tensor([1, 1])]; tensor value_83_pad_0 = const()[name = tensor("value_83_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_83_dilations_0 = const()[name = tensor("value_83_dilations_0"), val = tensor([1, 1])]; tensor value_83_groups_0 = const()[name = tensor("value_83_groups_0"), val = tensor(1)]; tensor layers_20_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_20_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1203086016)))]; tensor layers_20_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_20_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1206362880)))]; tensor value_83_cast_fp16 = conv(bias = layers_20_encoder_attn_v_proj_bias_to_fp16, dilations = value_83_dilations_0, groups = value_83_groups_0, pad = value_83_pad_0, pad_type = value_83_pad_type_0, strides = value_83_strides_0, weight = layers_20_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_83_cast_fp16")]; tensor var_4731 = const()[name = tensor("op_4731"), val = tensor([1, 20, 64, 1])]; tensor mh_q_83_cast_fp16 = reshape(shape = var_4731, x = query_83_cast_fp16)[name = tensor("mh_q_83_cast_fp16")]; tensor var_4733_to_fp16 = const()[name = tensor("op_4733_to_fp16"), val = tensor(0x1p-3)]; tensor var_4734_cast_fp16 = mul(x = mh_q_83_cast_fp16, y = var_4733_to_fp16)[name = tensor("op_4734_cast_fp16")]; tensor var_4737 = const()[name = tensor("op_4737"), val = tensor([1, 20, 64, 1500])]; tensor var_4738_cast_fp16 = reshape(shape = var_4737, x = key_83_cast_fp16)[name = tensor("op_4738_cast_fp16")]; tensor mh_w_125_transpose_x_0 = const()[name = tensor("mh_w_125_transpose_x_0"), val = tensor(true)]; tensor mh_w_125_transpose_y_0 = const()[name = tensor("mh_w_125_transpose_y_0"), val = tensor(false)]; tensor mh_w_125_cast_fp16 = matmul(transpose_x = mh_w_125_transpose_x_0, transpose_y = mh_w_125_transpose_y_0, x = var_4734_cast_fp16, y = var_4738_cast_fp16)[name = tensor("mh_w_125_cast_fp16")]; tensor obj_293_cast_fp16 = softmax(axis = var_4580, x = mh_w_125_cast_fp16)[name = tensor("obj_293_cast_fp16")]; tensor var_4742 = const()[name = tensor("op_4742"), val = tensor([1, 20, 64, 1500])]; tensor var_4743_cast_fp16 = reshape(shape = var_4742, x = value_83_cast_fp16)[name = tensor("op_4743_cast_fp16")]; tensor attn_83_transpose_x_0 = const()[name = tensor("attn_83_transpose_x_0"), val = tensor(false)]; tensor attn_83_transpose_y_0 = const()[name = tensor("attn_83_transpose_y_0"), val = tensor(true)]; tensor attn_83_cast_fp16 = matmul(transpose_x = attn_83_transpose_x_0, transpose_y = attn_83_transpose_y_0, x = var_4743_cast_fp16, y = obj_293_cast_fp16)[name = tensor("attn_83_cast_fp16")]; tensor var_4746 = const()[name = tensor("op_4746"), val = tensor([1, 1280, 1, 1])]; tensor input_203_cast_fp16 = reshape(shape = var_4746, x = attn_83_cast_fp16)[name = tensor("input_203_cast_fp16")]; tensor obj_291_pad_type_0 = const()[name = tensor("obj_291_pad_type_0"), val = tensor("valid")]; tensor obj_291_strides_0 = const()[name = tensor("obj_291_strides_0"), val = tensor([1, 1])]; tensor obj_291_pad_0 = const()[name = tensor("obj_291_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_291_dilations_0 = const()[name = tensor("obj_291_dilations_0"), val = tensor([1, 1])]; tensor obj_291_groups_0 = const()[name = tensor("obj_291_groups_0"), val = tensor(1)]; tensor layers_20_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_20_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1206365504)))]; tensor layers_20_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_20_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1209642368)))]; tensor obj_291_cast_fp16 = conv(bias = layers_20_encoder_attn_o_proj_bias_to_fp16, dilations = obj_291_dilations_0, groups = obj_291_groups_0, pad = obj_291_pad_0, pad_type = obj_291_pad_type_0, strides = obj_291_strides_0, weight = layers_20_encoder_attn_o_proj_weight_to_fp16, x = input_203_cast_fp16)[name = tensor("obj_291_cast_fp16")]; tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = obj_291_cast_fp16)[name = tensor("inputs_125_cast_fp16")]; tensor out_125_axes_0 = const()[name = tensor("out_125_axes_0"), val = tensor([1])]; tensor var_4764_to_fp16 = const()[name = tensor("op_4764_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_4764_to_fp16, x = inputs_125_cast_fp16)[name = tensor("out_125_cast_fp16")]; tensor input_205_gamma_0_to_fp16 = const()[name = tensor("input_205_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1209644992)))]; tensor input_205_beta_0_to_fp16 = const()[name = tensor("input_205_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1209647616)))]; tensor input_205_epsilon_0_to_fp16 = const()[name = tensor("input_205_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_205_cast_fp16 = batch_norm(beta = input_205_beta_0_to_fp16, epsilon = input_205_epsilon_0_to_fp16, gamma = input_205_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_125_cast_fp16)[name = tensor("input_205_cast_fp16")]; tensor input_207_pad_type_0 = const()[name = tensor("input_207_pad_type_0"), val = tensor("valid")]; tensor input_207_strides_0 = const()[name = tensor("input_207_strides_0"), val = tensor([1, 1])]; tensor input_207_pad_0 = const()[name = tensor("input_207_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_207_dilations_0 = const()[name = tensor("input_207_dilations_0"), val = tensor([1, 1])]; tensor input_207_groups_0 = const()[name = tensor("input_207_groups_0"), val = tensor(1)]; tensor layers_20_fc1_weight_to_fp16 = const()[name = tensor("layers_20_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1209650240)))]; tensor layers_20_fc1_bias_to_fp16 = const()[name = tensor("layers_20_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1222757504)))]; tensor input_207_cast_fp16 = conv(bias = layers_20_fc1_bias_to_fp16, dilations = input_207_dilations_0, groups = input_207_groups_0, pad = input_207_pad_0, pad_type = input_207_pad_type_0, strides = input_207_strides_0, weight = layers_20_fc1_weight_to_fp16, x = input_205_cast_fp16)[name = tensor("input_207_cast_fp16")]; tensor input_209_mode_0 = const()[name = tensor("input_209_mode_0"), val = tensor("EXACT")]; tensor input_209_cast_fp16 = gelu(mode = input_209_mode_0, x = input_207_cast_fp16)[name = tensor("input_209_cast_fp16")]; tensor hidden_states_43_pad_type_0 = const()[name = tensor("hidden_states_43_pad_type_0"), val = tensor("valid")]; tensor hidden_states_43_strides_0 = const()[name = tensor("hidden_states_43_strides_0"), val = tensor([1, 1])]; tensor hidden_states_43_pad_0 = const()[name = tensor("hidden_states_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_43_dilations_0 = const()[name = tensor("hidden_states_43_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_43_groups_0 = const()[name = tensor("hidden_states_43_groups_0"), val = tensor(1)]; tensor layers_20_fc2_weight_to_fp16 = const()[name = tensor("layers_20_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1222767808)))]; tensor layers_20_fc2_bias_to_fp16 = const()[name = tensor("layers_20_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1235875072)))]; tensor hidden_states_43_cast_fp16 = conv(bias = layers_20_fc2_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = layers_20_fc2_weight_to_fp16, x = input_209_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = hidden_states_43_cast_fp16)[name = tensor("inputs_127_cast_fp16")]; tensor var_4799 = const()[name = tensor("op_4799"), val = tensor(3)]; tensor out_127_axes_0 = const()[name = tensor("out_127_axes_0"), val = tensor([1])]; tensor var_4824_to_fp16 = const()[name = tensor("op_4824_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_4824_to_fp16, x = inputs_127_cast_fp16)[name = tensor("out_127_cast_fp16")]; tensor obj_295_gamma_0_to_fp16 = const()[name = tensor("obj_295_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1235877696)))]; tensor obj_295_beta_0_to_fp16 = const()[name = tensor("obj_295_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1235880320)))]; tensor obj_295_epsilon_0_to_fp16 = const()[name = tensor("obj_295_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_295_cast_fp16 = batch_norm(beta = obj_295_beta_0_to_fp16, epsilon = obj_295_epsilon_0_to_fp16, gamma = obj_295_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_127_cast_fp16)[name = tensor("obj_295_cast_fp16")]; tensor query_85_pad_type_0 = const()[name = tensor("query_85_pad_type_0"), val = tensor("valid")]; tensor query_85_strides_0 = const()[name = tensor("query_85_strides_0"), val = tensor([1, 1])]; tensor query_85_pad_0 = const()[name = tensor("query_85_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_85_dilations_0 = const()[name = tensor("query_85_dilations_0"), val = tensor([1, 1])]; tensor query_85_groups_0 = const()[name = tensor("query_85_groups_0"), val = tensor(1)]; tensor layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1235882944)))]; tensor layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1239159808)))]; tensor query_85_cast_fp16 = conv(bias = layers_21_self_attn_q_proj_bias_to_fp16, dilations = query_85_dilations_0, groups = query_85_groups_0, pad = query_85_pad_0, pad_type = query_85_pad_type_0, strides = query_85_strides_0, weight = layers_21_self_attn_q_proj_weight_to_fp16, x = obj_295_cast_fp16)[name = tensor("query_85_cast_fp16")]; tensor current_key_43_pad_type_0 = const()[name = tensor("current_key_43_pad_type_0"), val = tensor("valid")]; tensor current_key_43_strides_0 = const()[name = tensor("current_key_43_strides_0"), val = tensor([1, 1])]; tensor current_key_43_pad_0 = const()[name = tensor("current_key_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_43_dilations_0 = const()[name = tensor("current_key_43_dilations_0"), val = tensor([1, 1])]; tensor current_key_43_groups_0 = const()[name = tensor("current_key_43_groups_0"), val = tensor(1)]; tensor layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1239162432)))]; tensor current_key_43_cast_fp16 = conv(dilations = current_key_43_dilations_0, groups = current_key_43_groups_0, pad = current_key_43_pad_0, pad_type = current_key_43_pad_type_0, strides = current_key_43_strides_0, weight = layers_21_self_attn_k_proj_weight_to_fp16, x = obj_295_cast_fp16)[name = tensor("current_key_43_cast_fp16")]; tensor current_value_43_pad_type_0 = const()[name = tensor("current_value_43_pad_type_0"), val = tensor("valid")]; tensor current_value_43_strides_0 = const()[name = tensor("current_value_43_strides_0"), val = tensor([1, 1])]; tensor current_value_43_pad_0 = const()[name = tensor("current_value_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_43_dilations_0 = const()[name = tensor("current_value_43_dilations_0"), val = tensor([1, 1])]; tensor current_value_43_groups_0 = const()[name = tensor("current_value_43_groups_0"), val = tensor(1)]; tensor layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1242439296)))]; tensor layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1245716160)))]; tensor current_value_43_cast_fp16 = conv(bias = layers_21_self_attn_v_proj_bias_to_fp16, dilations = current_value_43_dilations_0, groups = current_value_43_groups_0, pad = current_value_43_pad_0, pad_type = current_value_43_pad_type_0, strides = current_value_43_strides_0, weight = layers_21_self_attn_v_proj_weight_to_fp16, x = obj_295_cast_fp16)[name = tensor("current_value_43_cast_fp16")]; tensor var_4863_cast_fp16 = mul(x = var_103_cast_fp16_21, y = var_239_cast_fp16)[name = tensor("op_4863_cast_fp16")]; tensor var_4864_cast_fp16 = mul(x = current_key_43_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_4864_cast_fp16")]; tensor key_85_cast_fp16 = add(x = var_4863_cast_fp16, y = var_4864_cast_fp16)[name = tensor("key_85_cast_fp16")]; tensor var_4867_cast_fp16 = mul(x = var_138_cast_fp16_21, y = var_239_cast_fp16)[name = tensor("op_4867_cast_fp16")]; tensor var_4868_cast_fp16 = mul(x = current_value_43_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_4868_cast_fp16")]; tensor value_85_cast_fp16 = add(x = var_4867_cast_fp16, y = var_4868_cast_fp16)[name = tensor("value_85_cast_fp16")]; tensor var_4872 = const()[name = tensor("op_4872"), val = tensor([1, 20, 64, 1])]; tensor mh_q_85_cast_fp16 = reshape(shape = var_4872, x = query_85_cast_fp16)[name = tensor("mh_q_85_cast_fp16")]; tensor var_4874_to_fp16 = const()[name = tensor("op_4874_to_fp16"), val = tensor(0x1p-3)]; tensor var_4875_cast_fp16 = mul(x = mh_q_85_cast_fp16, y = var_4874_to_fp16)[name = tensor("op_4875_cast_fp16")]; tensor var_4878 = const()[name = tensor("op_4878"), val = tensor([1, 20, 64, 448])]; tensor var_4879_cast_fp16 = reshape(shape = var_4878, x = key_85_cast_fp16)[name = tensor("op_4879_cast_fp16")]; tensor mh_w_127_transpose_x_0 = const()[name = tensor("mh_w_127_transpose_x_0"), val = tensor(true)]; tensor mh_w_127_transpose_y_0 = const()[name = tensor("mh_w_127_transpose_y_0"), val = tensor(false)]; tensor mh_w_127_cast_fp16 = matmul(transpose_x = mh_w_127_transpose_x_0, transpose_y = mh_w_127_transpose_y_0, x = var_4875_cast_fp16, y = var_4879_cast_fp16)[name = tensor("mh_w_127_cast_fp16")]; tensor mh_w_129_cast_fp16 = add(x = mh_w_127_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_129_cast_fp16")]; tensor var_4887_cast_fp16 = softmax(axis = var_4799, x = mh_w_129_cast_fp16)[name = tensor("op_4887_cast_fp16")]; tensor var_4888 = const()[name = tensor("op_4888"), val = tensor([1, 20, 64, 448])]; tensor var_4889_cast_fp16 = reshape(shape = var_4888, x = value_85_cast_fp16)[name = tensor("op_4889_cast_fp16")]; tensor attn_85_transpose_x_0 = const()[name = tensor("attn_85_transpose_x_0"), val = tensor(false)]; tensor attn_85_transpose_y_0 = const()[name = tensor("attn_85_transpose_y_0"), val = tensor(true)]; tensor attn_85_cast_fp16 = matmul(transpose_x = attn_85_transpose_x_0, transpose_y = attn_85_transpose_y_0, x = var_4889_cast_fp16, y = var_4887_cast_fp16)[name = tensor("attn_85_cast_fp16")]; tensor var_4892 = const()[name = tensor("op_4892"), val = tensor([1, 1280, 1, 1])]; tensor input_211_cast_fp16 = reshape(shape = var_4892, x = attn_85_cast_fp16)[name = tensor("input_211_cast_fp16")]; tensor obj_301_pad_type_0 = const()[name = tensor("obj_301_pad_type_0"), val = tensor("valid")]; tensor obj_301_strides_0 = const()[name = tensor("obj_301_strides_0"), val = tensor([1, 1])]; tensor obj_301_pad_0 = const()[name = tensor("obj_301_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_301_dilations_0 = const()[name = tensor("obj_301_dilations_0"), val = tensor([1, 1])]; tensor obj_301_groups_0 = const()[name = tensor("obj_301_groups_0"), val = tensor(1)]; tensor layers_21_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_21_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1245718784)))]; tensor layers_21_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_21_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1248995648)))]; tensor obj_301_cast_fp16 = conv(bias = layers_21_self_attn_o_proj_bias_to_fp16, dilations = obj_301_dilations_0, groups = obj_301_groups_0, pad = obj_301_pad_0, pad_type = obj_301_pad_type_0, strides = obj_301_strides_0, weight = layers_21_self_attn_o_proj_weight_to_fp16, x = input_211_cast_fp16)[name = tensor("obj_301_cast_fp16")]; tensor inputs_129_cast_fp16 = add(x = inputs_127_cast_fp16, y = obj_301_cast_fp16)[name = tensor("inputs_129_cast_fp16")]; tensor out_129_axes_0 = const()[name = tensor("out_129_axes_0"), val = tensor([1])]; tensor var_4914_to_fp16 = const()[name = tensor("op_4914_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_129_cast_fp16 = layer_norm(axes = out_129_axes_0, epsilon = var_4914_to_fp16, x = inputs_129_cast_fp16)[name = tensor("out_129_cast_fp16")]; tensor obj_303_gamma_0_to_fp16 = const()[name = tensor("obj_303_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1248998272)))]; tensor obj_303_beta_0_to_fp16 = const()[name = tensor("obj_303_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1249000896)))]; tensor obj_303_epsilon_0_to_fp16 = const()[name = tensor("obj_303_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_303_cast_fp16 = batch_norm(beta = obj_303_beta_0_to_fp16, epsilon = obj_303_epsilon_0_to_fp16, gamma = obj_303_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_129_cast_fp16)[name = tensor("obj_303_cast_fp16")]; tensor query_87_pad_type_0 = const()[name = tensor("query_87_pad_type_0"), val = tensor("valid")]; tensor query_87_strides_0 = const()[name = tensor("query_87_strides_0"), val = tensor([1, 1])]; tensor query_87_pad_0 = const()[name = tensor("query_87_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_87_dilations_0 = const()[name = tensor("query_87_dilations_0"), val = tensor([1, 1])]; tensor query_87_groups_0 = const()[name = tensor("query_87_groups_0"), val = tensor(1)]; tensor layers_21_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_21_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1249003520)))]; tensor layers_21_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_21_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1252280384)))]; tensor query_87_cast_fp16 = conv(bias = layers_21_encoder_attn_q_proj_bias_to_fp16, dilations = query_87_dilations_0, groups = query_87_groups_0, pad = query_87_pad_0, pad_type = query_87_pad_type_0, strides = query_87_strides_0, weight = layers_21_encoder_attn_q_proj_weight_to_fp16, x = obj_303_cast_fp16)[name = tensor("query_87_cast_fp16")]; tensor key_87_pad_type_0 = const()[name = tensor("key_87_pad_type_0"), val = tensor("valid")]; tensor key_87_strides_0 = const()[name = tensor("key_87_strides_0"), val = tensor([1, 1])]; tensor key_87_pad_0 = const()[name = tensor("key_87_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_87_dilations_0 = const()[name = tensor("key_87_dilations_0"), val = tensor([1, 1])]; tensor key_87_groups_0 = const()[name = tensor("key_87_groups_0"), val = tensor(1)]; tensor layers_21_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_21_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1252283008)))]; tensor key_87_cast_fp16 = conv(dilations = key_87_dilations_0, groups = key_87_groups_0, pad = key_87_pad_0, pad_type = key_87_pad_type_0, strides = key_87_strides_0, weight = layers_21_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_87_cast_fp16")]; tensor value_87_pad_type_0 = const()[name = tensor("value_87_pad_type_0"), val = tensor("valid")]; tensor value_87_strides_0 = const()[name = tensor("value_87_strides_0"), val = tensor([1, 1])]; tensor value_87_pad_0 = const()[name = tensor("value_87_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_87_dilations_0 = const()[name = tensor("value_87_dilations_0"), val = tensor([1, 1])]; tensor value_87_groups_0 = const()[name = tensor("value_87_groups_0"), val = tensor(1)]; tensor layers_21_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_21_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1255559872)))]; tensor layers_21_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_21_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1258836736)))]; tensor value_87_cast_fp16 = conv(bias = layers_21_encoder_attn_v_proj_bias_to_fp16, dilations = value_87_dilations_0, groups = value_87_groups_0, pad = value_87_pad_0, pad_type = value_87_pad_type_0, strides = value_87_strides_0, weight = layers_21_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_87_cast_fp16")]; tensor var_4950 = const()[name = tensor("op_4950"), val = tensor([1, 20, 64, 1])]; tensor mh_q_87_cast_fp16 = reshape(shape = var_4950, x = query_87_cast_fp16)[name = tensor("mh_q_87_cast_fp16")]; tensor var_4952_to_fp16 = const()[name = tensor("op_4952_to_fp16"), val = tensor(0x1p-3)]; tensor var_4953_cast_fp16 = mul(x = mh_q_87_cast_fp16, y = var_4952_to_fp16)[name = tensor("op_4953_cast_fp16")]; tensor var_4956 = const()[name = tensor("op_4956"), val = tensor([1, 20, 64, 1500])]; tensor var_4957_cast_fp16 = reshape(shape = var_4956, x = key_87_cast_fp16)[name = tensor("op_4957_cast_fp16")]; tensor mh_w_131_transpose_x_0 = const()[name = tensor("mh_w_131_transpose_x_0"), val = tensor(true)]; tensor mh_w_131_transpose_y_0 = const()[name = tensor("mh_w_131_transpose_y_0"), val = tensor(false)]; tensor mh_w_131_cast_fp16 = matmul(transpose_x = mh_w_131_transpose_x_0, transpose_y = mh_w_131_transpose_y_0, x = var_4953_cast_fp16, y = var_4957_cast_fp16)[name = tensor("mh_w_131_cast_fp16")]; tensor obj_307_cast_fp16 = softmax(axis = var_4799, x = mh_w_131_cast_fp16)[name = tensor("obj_307_cast_fp16")]; tensor var_4961 = const()[name = tensor("op_4961"), val = tensor([1, 20, 64, 1500])]; tensor var_4962_cast_fp16 = reshape(shape = var_4961, x = value_87_cast_fp16)[name = tensor("op_4962_cast_fp16")]; tensor attn_87_transpose_x_0 = const()[name = tensor("attn_87_transpose_x_0"), val = tensor(false)]; tensor attn_87_transpose_y_0 = const()[name = tensor("attn_87_transpose_y_0"), val = tensor(true)]; tensor attn_87_cast_fp16 = matmul(transpose_x = attn_87_transpose_x_0, transpose_y = attn_87_transpose_y_0, x = var_4962_cast_fp16, y = obj_307_cast_fp16)[name = tensor("attn_87_cast_fp16")]; tensor var_4965 = const()[name = tensor("op_4965"), val = tensor([1, 1280, 1, 1])]; tensor input_213_cast_fp16 = reshape(shape = var_4965, x = attn_87_cast_fp16)[name = tensor("input_213_cast_fp16")]; tensor obj_305_pad_type_0 = const()[name = tensor("obj_305_pad_type_0"), val = tensor("valid")]; tensor obj_305_strides_0 = const()[name = tensor("obj_305_strides_0"), val = tensor([1, 1])]; tensor obj_305_pad_0 = const()[name = tensor("obj_305_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_305_dilations_0 = const()[name = tensor("obj_305_dilations_0"), val = tensor([1, 1])]; tensor obj_305_groups_0 = const()[name = tensor("obj_305_groups_0"), val = tensor(1)]; tensor layers_21_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_21_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1258839360)))]; tensor layers_21_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_21_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1262116224)))]; tensor obj_305_cast_fp16 = conv(bias = layers_21_encoder_attn_o_proj_bias_to_fp16, dilations = obj_305_dilations_0, groups = obj_305_groups_0, pad = obj_305_pad_0, pad_type = obj_305_pad_type_0, strides = obj_305_strides_0, weight = layers_21_encoder_attn_o_proj_weight_to_fp16, x = input_213_cast_fp16)[name = tensor("obj_305_cast_fp16")]; tensor inputs_131_cast_fp16 = add(x = inputs_129_cast_fp16, y = obj_305_cast_fp16)[name = tensor("inputs_131_cast_fp16")]; tensor out_131_axes_0 = const()[name = tensor("out_131_axes_0"), val = tensor([1])]; tensor var_4986_to_fp16 = const()[name = tensor("op_4986_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_131_cast_fp16 = layer_norm(axes = out_131_axes_0, epsilon = var_4986_to_fp16, x = inputs_131_cast_fp16)[name = tensor("out_131_cast_fp16")]; tensor input_215_gamma_0_to_fp16 = const()[name = tensor("input_215_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1262118848)))]; tensor input_215_beta_0_to_fp16 = const()[name = tensor("input_215_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1262121472)))]; tensor input_215_epsilon_0_to_fp16 = const()[name = tensor("input_215_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_215_cast_fp16 = batch_norm(beta = input_215_beta_0_to_fp16, epsilon = input_215_epsilon_0_to_fp16, gamma = input_215_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_131_cast_fp16)[name = tensor("input_215_cast_fp16")]; tensor input_217_pad_type_0 = const()[name = tensor("input_217_pad_type_0"), val = tensor("valid")]; tensor input_217_strides_0 = const()[name = tensor("input_217_strides_0"), val = tensor([1, 1])]; tensor input_217_pad_0 = const()[name = tensor("input_217_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_217_dilations_0 = const()[name = tensor("input_217_dilations_0"), val = tensor([1, 1])]; tensor input_217_groups_0 = const()[name = tensor("input_217_groups_0"), val = tensor(1)]; tensor layers_21_fc1_weight_to_fp16 = const()[name = tensor("layers_21_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1262124096)))]; tensor layers_21_fc1_bias_to_fp16 = const()[name = tensor("layers_21_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1275231360)))]; tensor input_217_cast_fp16 = conv(bias = layers_21_fc1_bias_to_fp16, dilations = input_217_dilations_0, groups = input_217_groups_0, pad = input_217_pad_0, pad_type = input_217_pad_type_0, strides = input_217_strides_0, weight = layers_21_fc1_weight_to_fp16, x = input_215_cast_fp16)[name = tensor("input_217_cast_fp16")]; tensor input_219_mode_0 = const()[name = tensor("input_219_mode_0"), val = tensor("EXACT")]; tensor input_219_cast_fp16 = gelu(mode = input_219_mode_0, x = input_217_cast_fp16)[name = tensor("input_219_cast_fp16")]; tensor hidden_states_45_pad_type_0 = const()[name = tensor("hidden_states_45_pad_type_0"), val = tensor("valid")]; tensor hidden_states_45_strides_0 = const()[name = tensor("hidden_states_45_strides_0"), val = tensor([1, 1])]; tensor hidden_states_45_pad_0 = const()[name = tensor("hidden_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_45_dilations_0 = const()[name = tensor("hidden_states_45_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_45_groups_0 = const()[name = tensor("hidden_states_45_groups_0"), val = tensor(1)]; tensor layers_21_fc2_weight_to_fp16 = const()[name = tensor("layers_21_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1275241664)))]; tensor layers_21_fc2_bias_to_fp16 = const()[name = tensor("layers_21_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1288348928)))]; tensor hidden_states_45_cast_fp16 = conv(bias = layers_21_fc2_bias_to_fp16, dilations = hidden_states_45_dilations_0, groups = hidden_states_45_groups_0, pad = hidden_states_45_pad_0, pad_type = hidden_states_45_pad_type_0, strides = hidden_states_45_strides_0, weight = layers_21_fc2_weight_to_fp16, x = input_219_cast_fp16)[name = tensor("hidden_states_45_cast_fp16")]; tensor inputs_133_cast_fp16 = add(x = inputs_131_cast_fp16, y = hidden_states_45_cast_fp16)[name = tensor("inputs_133_cast_fp16")]; tensor var_5022 = const()[name = tensor("op_5022"), val = tensor(3)]; tensor out_133_axes_0 = const()[name = tensor("out_133_axes_0"), val = tensor([1])]; tensor var_5047_to_fp16 = const()[name = tensor("op_5047_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_133_cast_fp16 = layer_norm(axes = out_133_axes_0, epsilon = var_5047_to_fp16, x = inputs_133_cast_fp16)[name = tensor("out_133_cast_fp16")]; tensor obj_309_gamma_0_to_fp16 = const()[name = tensor("obj_309_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1288351552)))]; tensor obj_309_beta_0_to_fp16 = const()[name = tensor("obj_309_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1288354176)))]; tensor obj_309_epsilon_0_to_fp16 = const()[name = tensor("obj_309_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_309_cast_fp16 = batch_norm(beta = obj_309_beta_0_to_fp16, epsilon = obj_309_epsilon_0_to_fp16, gamma = obj_309_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_133_cast_fp16)[name = tensor("obj_309_cast_fp16")]; tensor query_89_pad_type_0 = const()[name = tensor("query_89_pad_type_0"), val = tensor("valid")]; tensor query_89_strides_0 = const()[name = tensor("query_89_strides_0"), val = tensor([1, 1])]; tensor query_89_pad_0 = const()[name = tensor("query_89_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_89_dilations_0 = const()[name = tensor("query_89_dilations_0"), val = tensor([1, 1])]; tensor query_89_groups_0 = const()[name = tensor("query_89_groups_0"), val = tensor(1)]; tensor layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1288356800)))]; tensor layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1291633664)))]; tensor query_89_cast_fp16 = conv(bias = layers_22_self_attn_q_proj_bias_to_fp16, dilations = query_89_dilations_0, groups = query_89_groups_0, pad = query_89_pad_0, pad_type = query_89_pad_type_0, strides = query_89_strides_0, weight = layers_22_self_attn_q_proj_weight_to_fp16, x = obj_309_cast_fp16)[name = tensor("query_89_cast_fp16")]; tensor current_key_45_pad_type_0 = const()[name = tensor("current_key_45_pad_type_0"), val = tensor("valid")]; tensor current_key_45_strides_0 = const()[name = tensor("current_key_45_strides_0"), val = tensor([1, 1])]; tensor current_key_45_pad_0 = const()[name = tensor("current_key_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_45_dilations_0 = const()[name = tensor("current_key_45_dilations_0"), val = tensor([1, 1])]; tensor current_key_45_groups_0 = const()[name = tensor("current_key_45_groups_0"), val = tensor(1)]; tensor layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1291636288)))]; tensor current_key_45_cast_fp16 = conv(dilations = current_key_45_dilations_0, groups = current_key_45_groups_0, pad = current_key_45_pad_0, pad_type = current_key_45_pad_type_0, strides = current_key_45_strides_0, weight = layers_22_self_attn_k_proj_weight_to_fp16, x = obj_309_cast_fp16)[name = tensor("current_key_45_cast_fp16")]; tensor current_value_45_pad_type_0 = const()[name = tensor("current_value_45_pad_type_0"), val = tensor("valid")]; tensor current_value_45_strides_0 = const()[name = tensor("current_value_45_strides_0"), val = tensor([1, 1])]; tensor current_value_45_pad_0 = const()[name = tensor("current_value_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_45_dilations_0 = const()[name = tensor("current_value_45_dilations_0"), val = tensor([1, 1])]; tensor current_value_45_groups_0 = const()[name = tensor("current_value_45_groups_0"), val = tensor(1)]; tensor layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1294913152)))]; tensor layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1298190016)))]; tensor current_value_45_cast_fp16 = conv(bias = layers_22_self_attn_v_proj_bias_to_fp16, dilations = current_value_45_dilations_0, groups = current_value_45_groups_0, pad = current_value_45_pad_0, pad_type = current_value_45_pad_type_0, strides = current_value_45_strides_0, weight = layers_22_self_attn_v_proj_weight_to_fp16, x = obj_309_cast_fp16)[name = tensor("current_value_45_cast_fp16")]; tensor var_5086_cast_fp16 = mul(x = var_103_cast_fp16_22, y = var_239_cast_fp16)[name = tensor("op_5086_cast_fp16")]; tensor var_5087_cast_fp16 = mul(x = current_key_45_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5087_cast_fp16")]; tensor key_89_cast_fp16 = add(x = var_5086_cast_fp16, y = var_5087_cast_fp16)[name = tensor("key_89_cast_fp16")]; tensor var_5090_cast_fp16 = mul(x = var_138_cast_fp16_22, y = var_239_cast_fp16)[name = tensor("op_5090_cast_fp16")]; tensor var_5091_cast_fp16 = mul(x = current_value_45_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5091_cast_fp16")]; tensor value_89_cast_fp16 = add(x = var_5090_cast_fp16, y = var_5091_cast_fp16)[name = tensor("value_89_cast_fp16")]; tensor var_5095 = const()[name = tensor("op_5095"), val = tensor([1, 20, 64, 1])]; tensor mh_q_89_cast_fp16 = reshape(shape = var_5095, x = query_89_cast_fp16)[name = tensor("mh_q_89_cast_fp16")]; tensor var_5097_to_fp16 = const()[name = tensor("op_5097_to_fp16"), val = tensor(0x1p-3)]; tensor var_5098_cast_fp16 = mul(x = mh_q_89_cast_fp16, y = var_5097_to_fp16)[name = tensor("op_5098_cast_fp16")]; tensor var_5101 = const()[name = tensor("op_5101"), val = tensor([1, 20, 64, 448])]; tensor var_5102_cast_fp16 = reshape(shape = var_5101, x = key_89_cast_fp16)[name = tensor("op_5102_cast_fp16")]; tensor mh_w_133_transpose_x_0 = const()[name = tensor("mh_w_133_transpose_x_0"), val = tensor(true)]; tensor mh_w_133_transpose_y_0 = const()[name = tensor("mh_w_133_transpose_y_0"), val = tensor(false)]; tensor mh_w_133_cast_fp16 = matmul(transpose_x = mh_w_133_transpose_x_0, transpose_y = mh_w_133_transpose_y_0, x = var_5098_cast_fp16, y = var_5102_cast_fp16)[name = tensor("mh_w_133_cast_fp16")]; tensor mh_w_135_cast_fp16 = add(x = mh_w_133_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_135_cast_fp16")]; tensor var_5110_cast_fp16 = softmax(axis = var_5022, x = mh_w_135_cast_fp16)[name = tensor("op_5110_cast_fp16")]; tensor var_5111 = const()[name = tensor("op_5111"), val = tensor([1, 20, 64, 448])]; tensor var_5112_cast_fp16 = reshape(shape = var_5111, x = value_89_cast_fp16)[name = tensor("op_5112_cast_fp16")]; tensor attn_89_transpose_x_0 = const()[name = tensor("attn_89_transpose_x_0"), val = tensor(false)]; tensor attn_89_transpose_y_0 = const()[name = tensor("attn_89_transpose_y_0"), val = tensor(true)]; tensor attn_89_cast_fp16 = matmul(transpose_x = attn_89_transpose_x_0, transpose_y = attn_89_transpose_y_0, x = var_5112_cast_fp16, y = var_5110_cast_fp16)[name = tensor("attn_89_cast_fp16")]; tensor var_5115 = const()[name = tensor("op_5115"), val = tensor([1, 1280, 1, 1])]; tensor input_221_cast_fp16 = reshape(shape = var_5115, x = attn_89_cast_fp16)[name = tensor("input_221_cast_fp16")]; tensor obj_315_pad_type_0 = const()[name = tensor("obj_315_pad_type_0"), val = tensor("valid")]; tensor obj_315_strides_0 = const()[name = tensor("obj_315_strides_0"), val = tensor([1, 1])]; tensor obj_315_pad_0 = const()[name = tensor("obj_315_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_315_dilations_0 = const()[name = tensor("obj_315_dilations_0"), val = tensor([1, 1])]; tensor obj_315_groups_0 = const()[name = tensor("obj_315_groups_0"), val = tensor(1)]; tensor layers_22_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_22_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1298192640)))]; tensor layers_22_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_22_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1301469504)))]; tensor obj_315_cast_fp16 = conv(bias = layers_22_self_attn_o_proj_bias_to_fp16, dilations = obj_315_dilations_0, groups = obj_315_groups_0, pad = obj_315_pad_0, pad_type = obj_315_pad_type_0, strides = obj_315_strides_0, weight = layers_22_self_attn_o_proj_weight_to_fp16, x = input_221_cast_fp16)[name = tensor("obj_315_cast_fp16")]; tensor inputs_135_cast_fp16 = add(x = inputs_133_cast_fp16, y = obj_315_cast_fp16)[name = tensor("inputs_135_cast_fp16")]; tensor out_135_axes_0 = const()[name = tensor("out_135_axes_0"), val = tensor([1])]; tensor var_5137_to_fp16 = const()[name = tensor("op_5137_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_135_cast_fp16 = layer_norm(axes = out_135_axes_0, epsilon = var_5137_to_fp16, x = inputs_135_cast_fp16)[name = tensor("out_135_cast_fp16")]; tensor obj_317_gamma_0_to_fp16 = const()[name = tensor("obj_317_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1301472128)))]; tensor obj_317_beta_0_to_fp16 = const()[name = tensor("obj_317_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1301474752)))]; tensor obj_317_epsilon_0_to_fp16 = const()[name = tensor("obj_317_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_317_cast_fp16 = batch_norm(beta = obj_317_beta_0_to_fp16, epsilon = obj_317_epsilon_0_to_fp16, gamma = obj_317_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_135_cast_fp16)[name = tensor("obj_317_cast_fp16")]; tensor query_91_pad_type_0 = const()[name = tensor("query_91_pad_type_0"), val = tensor("valid")]; tensor query_91_strides_0 = const()[name = tensor("query_91_strides_0"), val = tensor([1, 1])]; tensor query_91_pad_0 = const()[name = tensor("query_91_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_91_dilations_0 = const()[name = tensor("query_91_dilations_0"), val = tensor([1, 1])]; tensor query_91_groups_0 = const()[name = tensor("query_91_groups_0"), val = tensor(1)]; tensor layers_22_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_22_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1301477376)))]; tensor layers_22_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_22_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1304754240)))]; tensor query_91_cast_fp16 = conv(bias = layers_22_encoder_attn_q_proj_bias_to_fp16, dilations = query_91_dilations_0, groups = query_91_groups_0, pad = query_91_pad_0, pad_type = query_91_pad_type_0, strides = query_91_strides_0, weight = layers_22_encoder_attn_q_proj_weight_to_fp16, x = obj_317_cast_fp16)[name = tensor("query_91_cast_fp16")]; tensor key_91_pad_type_0 = const()[name = tensor("key_91_pad_type_0"), val = tensor("valid")]; tensor key_91_strides_0 = const()[name = tensor("key_91_strides_0"), val = tensor([1, 1])]; tensor key_91_pad_0 = const()[name = tensor("key_91_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_91_dilations_0 = const()[name = tensor("key_91_dilations_0"), val = tensor([1, 1])]; tensor key_91_groups_0 = const()[name = tensor("key_91_groups_0"), val = tensor(1)]; tensor layers_22_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_22_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1304756864)))]; tensor key_91_cast_fp16 = conv(dilations = key_91_dilations_0, groups = key_91_groups_0, pad = key_91_pad_0, pad_type = key_91_pad_type_0, strides = key_91_strides_0, weight = layers_22_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_91_cast_fp16")]; tensor value_91_pad_type_0 = const()[name = tensor("value_91_pad_type_0"), val = tensor("valid")]; tensor value_91_strides_0 = const()[name = tensor("value_91_strides_0"), val = tensor([1, 1])]; tensor value_91_pad_0 = const()[name = tensor("value_91_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_91_dilations_0 = const()[name = tensor("value_91_dilations_0"), val = tensor([1, 1])]; tensor value_91_groups_0 = const()[name = tensor("value_91_groups_0"), val = tensor(1)]; tensor layers_22_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_22_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1308033728)))]; tensor layers_22_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_22_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1311310592)))]; tensor value_91_cast_fp16 = conv(bias = layers_22_encoder_attn_v_proj_bias_to_fp16, dilations = value_91_dilations_0, groups = value_91_groups_0, pad = value_91_pad_0, pad_type = value_91_pad_type_0, strides = value_91_strides_0, weight = layers_22_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_91_cast_fp16")]; tensor var_5173 = const()[name = tensor("op_5173"), val = tensor([1, 20, 64, 1])]; tensor mh_q_91_cast_fp16 = reshape(shape = var_5173, x = query_91_cast_fp16)[name = tensor("mh_q_91_cast_fp16")]; tensor var_5175_to_fp16 = const()[name = tensor("op_5175_to_fp16"), val = tensor(0x1p-3)]; tensor var_5176_cast_fp16 = mul(x = mh_q_91_cast_fp16, y = var_5175_to_fp16)[name = tensor("op_5176_cast_fp16")]; tensor var_5179 = const()[name = tensor("op_5179"), val = tensor([1, 20, 64, 1500])]; tensor var_5180_cast_fp16 = reshape(shape = var_5179, x = key_91_cast_fp16)[name = tensor("op_5180_cast_fp16")]; tensor mh_w_137_transpose_x_0 = const()[name = tensor("mh_w_137_transpose_x_0"), val = tensor(true)]; tensor mh_w_137_transpose_y_0 = const()[name = tensor("mh_w_137_transpose_y_0"), val = tensor(false)]; tensor mh_w_137_cast_fp16 = matmul(transpose_x = mh_w_137_transpose_x_0, transpose_y = mh_w_137_transpose_y_0, x = var_5176_cast_fp16, y = var_5180_cast_fp16)[name = tensor("mh_w_137_cast_fp16")]; tensor obj_321_cast_fp16 = softmax(axis = var_5022, x = mh_w_137_cast_fp16)[name = tensor("obj_321_cast_fp16")]; tensor var_5184 = const()[name = tensor("op_5184"), val = tensor([1, 20, 64, 1500])]; tensor var_5185_cast_fp16 = reshape(shape = var_5184, x = value_91_cast_fp16)[name = tensor("op_5185_cast_fp16")]; tensor attn_91_transpose_x_0 = const()[name = tensor("attn_91_transpose_x_0"), val = tensor(false)]; tensor attn_91_transpose_y_0 = const()[name = tensor("attn_91_transpose_y_0"), val = tensor(true)]; tensor attn_91_cast_fp16 = matmul(transpose_x = attn_91_transpose_x_0, transpose_y = attn_91_transpose_y_0, x = var_5185_cast_fp16, y = obj_321_cast_fp16)[name = tensor("attn_91_cast_fp16")]; tensor var_5188 = const()[name = tensor("op_5188"), val = tensor([1, 1280, 1, 1])]; tensor input_223_cast_fp16 = reshape(shape = var_5188, x = attn_91_cast_fp16)[name = tensor("input_223_cast_fp16")]; tensor obj_319_pad_type_0 = const()[name = tensor("obj_319_pad_type_0"), val = tensor("valid")]; tensor obj_319_strides_0 = const()[name = tensor("obj_319_strides_0"), val = tensor([1, 1])]; tensor obj_319_pad_0 = const()[name = tensor("obj_319_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_319_dilations_0 = const()[name = tensor("obj_319_dilations_0"), val = tensor([1, 1])]; tensor obj_319_groups_0 = const()[name = tensor("obj_319_groups_0"), val = tensor(1)]; tensor layers_22_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_22_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1311313216)))]; tensor layers_22_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_22_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1314590080)))]; tensor obj_319_cast_fp16 = conv(bias = layers_22_encoder_attn_o_proj_bias_to_fp16, dilations = obj_319_dilations_0, groups = obj_319_groups_0, pad = obj_319_pad_0, pad_type = obj_319_pad_type_0, strides = obj_319_strides_0, weight = layers_22_encoder_attn_o_proj_weight_to_fp16, x = input_223_cast_fp16)[name = tensor("obj_319_cast_fp16")]; tensor inputs_137_cast_fp16 = add(x = inputs_135_cast_fp16, y = obj_319_cast_fp16)[name = tensor("inputs_137_cast_fp16")]; tensor out_137_axes_0 = const()[name = tensor("out_137_axes_0"), val = tensor([1])]; tensor var_5209_to_fp16 = const()[name = tensor("op_5209_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_137_cast_fp16 = layer_norm(axes = out_137_axes_0, epsilon = var_5209_to_fp16, x = inputs_137_cast_fp16)[name = tensor("out_137_cast_fp16")]; tensor input_225_gamma_0_to_fp16 = const()[name = tensor("input_225_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1314592704)))]; tensor input_225_beta_0_to_fp16 = const()[name = tensor("input_225_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1314595328)))]; tensor input_225_epsilon_0_to_fp16 = const()[name = tensor("input_225_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_225_cast_fp16 = batch_norm(beta = input_225_beta_0_to_fp16, epsilon = input_225_epsilon_0_to_fp16, gamma = input_225_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_137_cast_fp16)[name = tensor("input_225_cast_fp16")]; tensor input_227_pad_type_0 = const()[name = tensor("input_227_pad_type_0"), val = tensor("valid")]; tensor input_227_strides_0 = const()[name = tensor("input_227_strides_0"), val = tensor([1, 1])]; tensor input_227_pad_0 = const()[name = tensor("input_227_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_227_dilations_0 = const()[name = tensor("input_227_dilations_0"), val = tensor([1, 1])]; tensor input_227_groups_0 = const()[name = tensor("input_227_groups_0"), val = tensor(1)]; tensor layers_22_fc1_weight_to_fp16 = const()[name = tensor("layers_22_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1314597952)))]; tensor layers_22_fc1_bias_to_fp16 = const()[name = tensor("layers_22_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1327705216)))]; tensor input_227_cast_fp16 = conv(bias = layers_22_fc1_bias_to_fp16, dilations = input_227_dilations_0, groups = input_227_groups_0, pad = input_227_pad_0, pad_type = input_227_pad_type_0, strides = input_227_strides_0, weight = layers_22_fc1_weight_to_fp16, x = input_225_cast_fp16)[name = tensor("input_227_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 = input_227_cast_fp16)[name = tensor("input_229_cast_fp16")]; tensor hidden_states_47_pad_type_0 = const()[name = tensor("hidden_states_47_pad_type_0"), val = tensor("valid")]; tensor hidden_states_47_strides_0 = const()[name = tensor("hidden_states_47_strides_0"), val = tensor([1, 1])]; tensor hidden_states_47_pad_0 = const()[name = tensor("hidden_states_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_47_dilations_0 = const()[name = tensor("hidden_states_47_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_47_groups_0 = const()[name = tensor("hidden_states_47_groups_0"), val = tensor(1)]; tensor layers_22_fc2_weight_to_fp16 = const()[name = tensor("layers_22_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1327715520)))]; tensor layers_22_fc2_bias_to_fp16 = const()[name = tensor("layers_22_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1340822784)))]; tensor hidden_states_47_cast_fp16 = conv(bias = layers_22_fc2_bias_to_fp16, dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = layers_22_fc2_weight_to_fp16, x = input_229_cast_fp16)[name = tensor("hidden_states_47_cast_fp16")]; tensor inputs_139_cast_fp16 = add(x = inputs_137_cast_fp16, y = hidden_states_47_cast_fp16)[name = tensor("inputs_139_cast_fp16")]; tensor var_5245 = const()[name = tensor("op_5245"), val = tensor(3)]; tensor out_139_axes_0 = const()[name = tensor("out_139_axes_0"), val = tensor([1])]; tensor var_5270_to_fp16 = const()[name = tensor("op_5270_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_139_cast_fp16 = layer_norm(axes = out_139_axes_0, epsilon = var_5270_to_fp16, x = inputs_139_cast_fp16)[name = tensor("out_139_cast_fp16")]; tensor obj_323_gamma_0_to_fp16 = const()[name = tensor("obj_323_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1340825408)))]; tensor obj_323_beta_0_to_fp16 = const()[name = tensor("obj_323_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1340828032)))]; tensor obj_323_epsilon_0_to_fp16 = const()[name = tensor("obj_323_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_323_cast_fp16 = batch_norm(beta = obj_323_beta_0_to_fp16, epsilon = obj_323_epsilon_0_to_fp16, gamma = obj_323_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_139_cast_fp16)[name = tensor("obj_323_cast_fp16")]; tensor query_93_pad_type_0 = const()[name = tensor("query_93_pad_type_0"), val = tensor("valid")]; tensor query_93_strides_0 = const()[name = tensor("query_93_strides_0"), val = tensor([1, 1])]; tensor query_93_pad_0 = const()[name = tensor("query_93_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_93_dilations_0 = const()[name = tensor("query_93_dilations_0"), val = tensor([1, 1])]; tensor query_93_groups_0 = const()[name = tensor("query_93_groups_0"), val = tensor(1)]; tensor layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1340830656)))]; tensor layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1344107520)))]; tensor query_93_cast_fp16 = conv(bias = layers_23_self_attn_q_proj_bias_to_fp16, dilations = query_93_dilations_0, groups = query_93_groups_0, pad = query_93_pad_0, pad_type = query_93_pad_type_0, strides = query_93_strides_0, weight = layers_23_self_attn_q_proj_weight_to_fp16, x = obj_323_cast_fp16)[name = tensor("query_93_cast_fp16")]; tensor current_key_47_pad_type_0 = const()[name = tensor("current_key_47_pad_type_0"), val = tensor("valid")]; tensor current_key_47_strides_0 = const()[name = tensor("current_key_47_strides_0"), val = tensor([1, 1])]; tensor current_key_47_pad_0 = const()[name = tensor("current_key_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_47_dilations_0 = const()[name = tensor("current_key_47_dilations_0"), val = tensor([1, 1])]; tensor current_key_47_groups_0 = const()[name = tensor("current_key_47_groups_0"), val = tensor(1)]; tensor layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1344110144)))]; tensor current_key_47_cast_fp16 = conv(dilations = current_key_47_dilations_0, groups = current_key_47_groups_0, pad = current_key_47_pad_0, pad_type = current_key_47_pad_type_0, strides = current_key_47_strides_0, weight = layers_23_self_attn_k_proj_weight_to_fp16, x = obj_323_cast_fp16)[name = tensor("current_key_47_cast_fp16")]; tensor current_value_47_pad_type_0 = const()[name = tensor("current_value_47_pad_type_0"), val = tensor("valid")]; tensor current_value_47_strides_0 = const()[name = tensor("current_value_47_strides_0"), val = tensor([1, 1])]; tensor current_value_47_pad_0 = const()[name = tensor("current_value_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_47_dilations_0 = const()[name = tensor("current_value_47_dilations_0"), val = tensor([1, 1])]; tensor current_value_47_groups_0 = const()[name = tensor("current_value_47_groups_0"), val = tensor(1)]; tensor layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1347387008)))]; tensor layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1350663872)))]; tensor current_value_47_cast_fp16 = conv(bias = layers_23_self_attn_v_proj_bias_to_fp16, dilations = current_value_47_dilations_0, groups = current_value_47_groups_0, pad = current_value_47_pad_0, pad_type = current_value_47_pad_type_0, strides = current_value_47_strides_0, weight = layers_23_self_attn_v_proj_weight_to_fp16, x = obj_323_cast_fp16)[name = tensor("current_value_47_cast_fp16")]; tensor var_5309_cast_fp16 = mul(x = var_103_cast_fp16_23, y = var_239_cast_fp16)[name = tensor("op_5309_cast_fp16")]; tensor var_5310_cast_fp16 = mul(x = current_key_47_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5310_cast_fp16")]; tensor key_93_cast_fp16 = add(x = var_5309_cast_fp16, y = var_5310_cast_fp16)[name = tensor("key_93_cast_fp16")]; tensor var_5313_cast_fp16 = mul(x = var_138_cast_fp16_23, y = var_239_cast_fp16)[name = tensor("op_5313_cast_fp16")]; tensor var_5314_cast_fp16 = mul(x = current_value_47_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5314_cast_fp16")]; tensor value_93_cast_fp16 = add(x = var_5313_cast_fp16, y = var_5314_cast_fp16)[name = tensor("value_93_cast_fp16")]; tensor var_5318 = const()[name = tensor("op_5318"), val = tensor([1, 20, 64, 1])]; tensor mh_q_93_cast_fp16 = reshape(shape = var_5318, x = query_93_cast_fp16)[name = tensor("mh_q_93_cast_fp16")]; tensor var_5320_to_fp16 = const()[name = tensor("op_5320_to_fp16"), val = tensor(0x1p-3)]; tensor var_5321_cast_fp16 = mul(x = mh_q_93_cast_fp16, y = var_5320_to_fp16)[name = tensor("op_5321_cast_fp16")]; tensor var_5324 = const()[name = tensor("op_5324"), val = tensor([1, 20, 64, 448])]; tensor var_5325_cast_fp16 = reshape(shape = var_5324, x = key_93_cast_fp16)[name = tensor("op_5325_cast_fp16")]; tensor mh_w_139_transpose_x_0 = const()[name = tensor("mh_w_139_transpose_x_0"), val = tensor(true)]; tensor mh_w_139_transpose_y_0 = const()[name = tensor("mh_w_139_transpose_y_0"), val = tensor(false)]; tensor mh_w_139_cast_fp16 = matmul(transpose_x = mh_w_139_transpose_x_0, transpose_y = mh_w_139_transpose_y_0, x = var_5321_cast_fp16, y = var_5325_cast_fp16)[name = tensor("mh_w_139_cast_fp16")]; tensor mh_w_141_cast_fp16 = add(x = mh_w_139_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_141_cast_fp16")]; tensor var_5333_cast_fp16 = softmax(axis = var_5245, x = mh_w_141_cast_fp16)[name = tensor("op_5333_cast_fp16")]; tensor var_5334 = const()[name = tensor("op_5334"), val = tensor([1, 20, 64, 448])]; tensor var_5335_cast_fp16 = reshape(shape = var_5334, x = value_93_cast_fp16)[name = tensor("op_5335_cast_fp16")]; tensor attn_93_transpose_x_0 = const()[name = tensor("attn_93_transpose_x_0"), val = tensor(false)]; tensor attn_93_transpose_y_0 = const()[name = tensor("attn_93_transpose_y_0"), val = tensor(true)]; tensor attn_93_cast_fp16 = matmul(transpose_x = attn_93_transpose_x_0, transpose_y = attn_93_transpose_y_0, x = var_5335_cast_fp16, y = var_5333_cast_fp16)[name = tensor("attn_93_cast_fp16")]; tensor var_5338 = const()[name = tensor("op_5338"), val = tensor([1, 1280, 1, 1])]; tensor input_231_cast_fp16 = reshape(shape = var_5338, x = attn_93_cast_fp16)[name = tensor("input_231_cast_fp16")]; tensor obj_329_pad_type_0 = const()[name = tensor("obj_329_pad_type_0"), val = tensor("valid")]; tensor obj_329_strides_0 = const()[name = tensor("obj_329_strides_0"), val = tensor([1, 1])]; tensor obj_329_pad_0 = const()[name = tensor("obj_329_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_329_dilations_0 = const()[name = tensor("obj_329_dilations_0"), val = tensor([1, 1])]; tensor obj_329_groups_0 = const()[name = tensor("obj_329_groups_0"), val = tensor(1)]; tensor layers_23_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_23_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1350666496)))]; tensor layers_23_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_23_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1353943360)))]; tensor obj_329_cast_fp16 = conv(bias = layers_23_self_attn_o_proj_bias_to_fp16, dilations = obj_329_dilations_0, groups = obj_329_groups_0, pad = obj_329_pad_0, pad_type = obj_329_pad_type_0, strides = obj_329_strides_0, weight = layers_23_self_attn_o_proj_weight_to_fp16, x = input_231_cast_fp16)[name = tensor("obj_329_cast_fp16")]; tensor inputs_141_cast_fp16 = add(x = inputs_139_cast_fp16, y = obj_329_cast_fp16)[name = tensor("inputs_141_cast_fp16")]; tensor out_141_axes_0 = const()[name = tensor("out_141_axes_0"), val = tensor([1])]; tensor var_5360_to_fp16 = const()[name = tensor("op_5360_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_141_cast_fp16 = layer_norm(axes = out_141_axes_0, epsilon = var_5360_to_fp16, x = inputs_141_cast_fp16)[name = tensor("out_141_cast_fp16")]; tensor obj_331_gamma_0_to_fp16 = const()[name = tensor("obj_331_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1353945984)))]; tensor obj_331_beta_0_to_fp16 = const()[name = tensor("obj_331_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1353948608)))]; tensor obj_331_epsilon_0_to_fp16 = const()[name = tensor("obj_331_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_331_cast_fp16 = batch_norm(beta = obj_331_beta_0_to_fp16, epsilon = obj_331_epsilon_0_to_fp16, gamma = obj_331_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_141_cast_fp16)[name = tensor("obj_331_cast_fp16")]; tensor query_95_pad_type_0 = const()[name = tensor("query_95_pad_type_0"), val = tensor("valid")]; tensor query_95_strides_0 = const()[name = tensor("query_95_strides_0"), val = tensor([1, 1])]; tensor query_95_pad_0 = const()[name = tensor("query_95_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_95_dilations_0 = const()[name = tensor("query_95_dilations_0"), val = tensor([1, 1])]; tensor query_95_groups_0 = const()[name = tensor("query_95_groups_0"), val = tensor(1)]; tensor layers_23_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_23_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1353951232)))]; tensor layers_23_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_23_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1357228096)))]; tensor query_95_cast_fp16 = conv(bias = layers_23_encoder_attn_q_proj_bias_to_fp16, dilations = query_95_dilations_0, groups = query_95_groups_0, pad = query_95_pad_0, pad_type = query_95_pad_type_0, strides = query_95_strides_0, weight = layers_23_encoder_attn_q_proj_weight_to_fp16, x = obj_331_cast_fp16)[name = tensor("query_95_cast_fp16")]; tensor key_95_pad_type_0 = const()[name = tensor("key_95_pad_type_0"), val = tensor("valid")]; tensor key_95_strides_0 = const()[name = tensor("key_95_strides_0"), val = tensor([1, 1])]; tensor key_95_pad_0 = const()[name = tensor("key_95_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_95_dilations_0 = const()[name = tensor("key_95_dilations_0"), val = tensor([1, 1])]; tensor key_95_groups_0 = const()[name = tensor("key_95_groups_0"), val = tensor(1)]; tensor layers_23_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_23_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1357230720)))]; tensor key_95_cast_fp16 = conv(dilations = key_95_dilations_0, groups = key_95_groups_0, pad = key_95_pad_0, pad_type = key_95_pad_type_0, strides = key_95_strides_0, weight = layers_23_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_95_cast_fp16")]; tensor value_95_pad_type_0 = const()[name = tensor("value_95_pad_type_0"), val = tensor("valid")]; tensor value_95_strides_0 = const()[name = tensor("value_95_strides_0"), val = tensor([1, 1])]; tensor value_95_pad_0 = const()[name = tensor("value_95_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_95_dilations_0 = const()[name = tensor("value_95_dilations_0"), val = tensor([1, 1])]; tensor value_95_groups_0 = const()[name = tensor("value_95_groups_0"), val = tensor(1)]; tensor layers_23_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_23_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1360507584)))]; tensor layers_23_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_23_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1363784448)))]; tensor value_95_cast_fp16 = conv(bias = layers_23_encoder_attn_v_proj_bias_to_fp16, dilations = value_95_dilations_0, groups = value_95_groups_0, pad = value_95_pad_0, pad_type = value_95_pad_type_0, strides = value_95_strides_0, weight = layers_23_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_95_cast_fp16")]; tensor var_5396 = const()[name = tensor("op_5396"), val = tensor([1, 20, 64, 1])]; tensor mh_q_95_cast_fp16 = reshape(shape = var_5396, x = query_95_cast_fp16)[name = tensor("mh_q_95_cast_fp16")]; tensor var_5398_to_fp16 = const()[name = tensor("op_5398_to_fp16"), val = tensor(0x1p-3)]; tensor var_5399_cast_fp16 = mul(x = mh_q_95_cast_fp16, y = var_5398_to_fp16)[name = tensor("op_5399_cast_fp16")]; tensor var_5402 = const()[name = tensor("op_5402"), val = tensor([1, 20, 64, 1500])]; tensor var_5403_cast_fp16 = reshape(shape = var_5402, x = key_95_cast_fp16)[name = tensor("op_5403_cast_fp16")]; tensor mh_w_143_transpose_x_0 = const()[name = tensor("mh_w_143_transpose_x_0"), val = tensor(true)]; tensor mh_w_143_transpose_y_0 = const()[name = tensor("mh_w_143_transpose_y_0"), val = tensor(false)]; tensor mh_w_143_cast_fp16 = matmul(transpose_x = mh_w_143_transpose_x_0, transpose_y = mh_w_143_transpose_y_0, x = var_5399_cast_fp16, y = var_5403_cast_fp16)[name = tensor("mh_w_143_cast_fp16")]; tensor obj_335_cast_fp16 = softmax(axis = var_5245, x = mh_w_143_cast_fp16)[name = tensor("obj_335_cast_fp16")]; tensor var_5407 = const()[name = tensor("op_5407"), val = tensor([1, 20, 64, 1500])]; tensor var_5408_cast_fp16 = reshape(shape = var_5407, x = value_95_cast_fp16)[name = tensor("op_5408_cast_fp16")]; tensor attn_95_transpose_x_0 = const()[name = tensor("attn_95_transpose_x_0"), val = tensor(false)]; tensor attn_95_transpose_y_0 = const()[name = tensor("attn_95_transpose_y_0"), val = tensor(true)]; tensor attn_95_cast_fp16 = matmul(transpose_x = attn_95_transpose_x_0, transpose_y = attn_95_transpose_y_0, x = var_5408_cast_fp16, y = obj_335_cast_fp16)[name = tensor("attn_95_cast_fp16")]; tensor var_5411 = const()[name = tensor("op_5411"), val = tensor([1, 1280, 1, 1])]; tensor input_233_cast_fp16 = reshape(shape = var_5411, x = attn_95_cast_fp16)[name = tensor("input_233_cast_fp16")]; tensor obj_333_pad_type_0 = const()[name = tensor("obj_333_pad_type_0"), val = tensor("valid")]; tensor obj_333_strides_0 = const()[name = tensor("obj_333_strides_0"), val = tensor([1, 1])]; tensor obj_333_pad_0 = const()[name = tensor("obj_333_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_333_dilations_0 = const()[name = tensor("obj_333_dilations_0"), val = tensor([1, 1])]; tensor obj_333_groups_0 = const()[name = tensor("obj_333_groups_0"), val = tensor(1)]; tensor layers_23_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_23_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1363787072)))]; tensor layers_23_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_23_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1367063936)))]; tensor obj_333_cast_fp16 = conv(bias = layers_23_encoder_attn_o_proj_bias_to_fp16, dilations = obj_333_dilations_0, groups = obj_333_groups_0, pad = obj_333_pad_0, pad_type = obj_333_pad_type_0, strides = obj_333_strides_0, weight = layers_23_encoder_attn_o_proj_weight_to_fp16, x = input_233_cast_fp16)[name = tensor("obj_333_cast_fp16")]; tensor inputs_143_cast_fp16 = add(x = inputs_141_cast_fp16, y = obj_333_cast_fp16)[name = tensor("inputs_143_cast_fp16")]; tensor out_143_axes_0 = const()[name = tensor("out_143_axes_0"), val = tensor([1])]; tensor var_5432_to_fp16 = const()[name = tensor("op_5432_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_143_cast_fp16 = layer_norm(axes = out_143_axes_0, epsilon = var_5432_to_fp16, x = inputs_143_cast_fp16)[name = tensor("out_143_cast_fp16")]; tensor input_235_gamma_0_to_fp16 = const()[name = tensor("input_235_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1367066560)))]; tensor input_235_beta_0_to_fp16 = const()[name = tensor("input_235_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1367069184)))]; tensor input_235_epsilon_0_to_fp16 = const()[name = tensor("input_235_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_235_cast_fp16 = batch_norm(beta = input_235_beta_0_to_fp16, epsilon = input_235_epsilon_0_to_fp16, gamma = input_235_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_143_cast_fp16)[name = tensor("input_235_cast_fp16")]; tensor input_237_pad_type_0 = const()[name = tensor("input_237_pad_type_0"), val = tensor("valid")]; tensor input_237_strides_0 = const()[name = tensor("input_237_strides_0"), val = tensor([1, 1])]; tensor input_237_pad_0 = const()[name = tensor("input_237_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_237_dilations_0 = const()[name = tensor("input_237_dilations_0"), val = tensor([1, 1])]; tensor input_237_groups_0 = const()[name = tensor("input_237_groups_0"), val = tensor(1)]; tensor layers_23_fc1_weight_to_fp16 = const()[name = tensor("layers_23_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1367071808)))]; tensor layers_23_fc1_bias_to_fp16 = const()[name = tensor("layers_23_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1380179072)))]; tensor input_237_cast_fp16 = conv(bias = layers_23_fc1_bias_to_fp16, dilations = input_237_dilations_0, groups = input_237_groups_0, pad = input_237_pad_0, pad_type = input_237_pad_type_0, strides = input_237_strides_0, weight = layers_23_fc1_weight_to_fp16, x = input_235_cast_fp16)[name = tensor("input_237_cast_fp16")]; tensor input_239_mode_0 = const()[name = tensor("input_239_mode_0"), val = tensor("EXACT")]; tensor input_239_cast_fp16 = gelu(mode = input_239_mode_0, x = input_237_cast_fp16)[name = tensor("input_239_cast_fp16")]; tensor hidden_states_49_pad_type_0 = const()[name = tensor("hidden_states_49_pad_type_0"), val = tensor("valid")]; tensor hidden_states_49_strides_0 = const()[name = tensor("hidden_states_49_strides_0"), val = tensor([1, 1])]; tensor hidden_states_49_pad_0 = const()[name = tensor("hidden_states_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_49_dilations_0 = const()[name = tensor("hidden_states_49_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_49_groups_0 = const()[name = tensor("hidden_states_49_groups_0"), val = tensor(1)]; tensor layers_23_fc2_weight_to_fp16 = const()[name = tensor("layers_23_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1380189376)))]; tensor layers_23_fc2_bias_to_fp16 = const()[name = tensor("layers_23_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1393296640)))]; tensor hidden_states_49_cast_fp16 = conv(bias = layers_23_fc2_bias_to_fp16, dilations = hidden_states_49_dilations_0, groups = hidden_states_49_groups_0, pad = hidden_states_49_pad_0, pad_type = hidden_states_49_pad_type_0, strides = hidden_states_49_strides_0, weight = layers_23_fc2_weight_to_fp16, x = input_239_cast_fp16)[name = tensor("hidden_states_49_cast_fp16")]; tensor inputs_145_cast_fp16 = add(x = inputs_143_cast_fp16, y = hidden_states_49_cast_fp16)[name = tensor("inputs_145_cast_fp16")]; tensor var_5468 = const()[name = tensor("op_5468"), val = tensor(3)]; tensor out_145_axes_0 = const()[name = tensor("out_145_axes_0"), val = tensor([1])]; tensor var_5493_to_fp16 = const()[name = tensor("op_5493_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_145_cast_fp16 = layer_norm(axes = out_145_axes_0, epsilon = var_5493_to_fp16, x = inputs_145_cast_fp16)[name = tensor("out_145_cast_fp16")]; tensor obj_337_gamma_0_to_fp16 = const()[name = tensor("obj_337_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1393299264)))]; tensor obj_337_beta_0_to_fp16 = const()[name = tensor("obj_337_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1393301888)))]; tensor obj_337_epsilon_0_to_fp16 = const()[name = tensor("obj_337_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_337_cast_fp16 = batch_norm(beta = obj_337_beta_0_to_fp16, epsilon = obj_337_epsilon_0_to_fp16, gamma = obj_337_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_145_cast_fp16)[name = tensor("obj_337_cast_fp16")]; tensor query_97_pad_type_0 = const()[name = tensor("query_97_pad_type_0"), val = tensor("valid")]; tensor query_97_strides_0 = const()[name = tensor("query_97_strides_0"), val = tensor([1, 1])]; tensor query_97_pad_0 = const()[name = tensor("query_97_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_97_dilations_0 = const()[name = tensor("query_97_dilations_0"), val = tensor([1, 1])]; tensor query_97_groups_0 = const()[name = tensor("query_97_groups_0"), val = tensor(1)]; tensor layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1393304512)))]; tensor layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1396581376)))]; tensor query_97_cast_fp16 = conv(bias = layers_24_self_attn_q_proj_bias_to_fp16, dilations = query_97_dilations_0, groups = query_97_groups_0, pad = query_97_pad_0, pad_type = query_97_pad_type_0, strides = query_97_strides_0, weight = layers_24_self_attn_q_proj_weight_to_fp16, x = obj_337_cast_fp16)[name = tensor("query_97_cast_fp16")]; tensor current_key_49_pad_type_0 = const()[name = tensor("current_key_49_pad_type_0"), val = tensor("valid")]; tensor current_key_49_strides_0 = const()[name = tensor("current_key_49_strides_0"), val = tensor([1, 1])]; tensor current_key_49_pad_0 = const()[name = tensor("current_key_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_49_dilations_0 = const()[name = tensor("current_key_49_dilations_0"), val = tensor([1, 1])]; tensor current_key_49_groups_0 = const()[name = tensor("current_key_49_groups_0"), val = tensor(1)]; tensor layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1396584000)))]; tensor current_key_49_cast_fp16 = conv(dilations = current_key_49_dilations_0, groups = current_key_49_groups_0, pad = current_key_49_pad_0, pad_type = current_key_49_pad_type_0, strides = current_key_49_strides_0, weight = layers_24_self_attn_k_proj_weight_to_fp16, x = obj_337_cast_fp16)[name = tensor("current_key_49_cast_fp16")]; tensor current_value_49_pad_type_0 = const()[name = tensor("current_value_49_pad_type_0"), val = tensor("valid")]; tensor current_value_49_strides_0 = const()[name = tensor("current_value_49_strides_0"), val = tensor([1, 1])]; tensor current_value_49_pad_0 = const()[name = tensor("current_value_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_49_dilations_0 = const()[name = tensor("current_value_49_dilations_0"), val = tensor([1, 1])]; tensor current_value_49_groups_0 = const()[name = tensor("current_value_49_groups_0"), val = tensor(1)]; tensor layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1399860864)))]; tensor layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1403137728)))]; tensor current_value_49_cast_fp16 = conv(bias = layers_24_self_attn_v_proj_bias_to_fp16, dilations = current_value_49_dilations_0, groups = current_value_49_groups_0, pad = current_value_49_pad_0, pad_type = current_value_49_pad_type_0, strides = current_value_49_strides_0, weight = layers_24_self_attn_v_proj_weight_to_fp16, x = obj_337_cast_fp16)[name = tensor("current_value_49_cast_fp16")]; tensor var_5532_cast_fp16 = mul(x = var_103_cast_fp16_24, y = var_239_cast_fp16)[name = tensor("op_5532_cast_fp16")]; tensor var_5533_cast_fp16 = mul(x = current_key_49_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5533_cast_fp16")]; tensor key_97_cast_fp16 = add(x = var_5532_cast_fp16, y = var_5533_cast_fp16)[name = tensor("key_97_cast_fp16")]; tensor var_5536_cast_fp16 = mul(x = var_138_cast_fp16_24, y = var_239_cast_fp16)[name = tensor("op_5536_cast_fp16")]; tensor var_5537_cast_fp16 = mul(x = current_value_49_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5537_cast_fp16")]; tensor value_97_cast_fp16 = add(x = var_5536_cast_fp16, y = var_5537_cast_fp16)[name = tensor("value_97_cast_fp16")]; tensor var_5541 = const()[name = tensor("op_5541"), val = tensor([1, 20, 64, 1])]; tensor mh_q_97_cast_fp16 = reshape(shape = var_5541, x = query_97_cast_fp16)[name = tensor("mh_q_97_cast_fp16")]; tensor var_5543_to_fp16 = const()[name = tensor("op_5543_to_fp16"), val = tensor(0x1p-3)]; tensor var_5544_cast_fp16 = mul(x = mh_q_97_cast_fp16, y = var_5543_to_fp16)[name = tensor("op_5544_cast_fp16")]; tensor var_5547 = const()[name = tensor("op_5547"), val = tensor([1, 20, 64, 448])]; tensor var_5548_cast_fp16 = reshape(shape = var_5547, x = key_97_cast_fp16)[name = tensor("op_5548_cast_fp16")]; tensor mh_w_145_transpose_x_0 = const()[name = tensor("mh_w_145_transpose_x_0"), val = tensor(true)]; tensor mh_w_145_transpose_y_0 = const()[name = tensor("mh_w_145_transpose_y_0"), val = tensor(false)]; tensor mh_w_145_cast_fp16 = matmul(transpose_x = mh_w_145_transpose_x_0, transpose_y = mh_w_145_transpose_y_0, x = var_5544_cast_fp16, y = var_5548_cast_fp16)[name = tensor("mh_w_145_cast_fp16")]; tensor mh_w_147_cast_fp16 = add(x = mh_w_145_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_147_cast_fp16")]; tensor var_5556_cast_fp16 = softmax(axis = var_5468, x = mh_w_147_cast_fp16)[name = tensor("op_5556_cast_fp16")]; tensor var_5557 = const()[name = tensor("op_5557"), val = tensor([1, 20, 64, 448])]; tensor var_5558_cast_fp16 = reshape(shape = var_5557, x = value_97_cast_fp16)[name = tensor("op_5558_cast_fp16")]; tensor attn_97_transpose_x_0 = const()[name = tensor("attn_97_transpose_x_0"), val = tensor(false)]; tensor attn_97_transpose_y_0 = const()[name = tensor("attn_97_transpose_y_0"), val = tensor(true)]; tensor attn_97_cast_fp16 = matmul(transpose_x = attn_97_transpose_x_0, transpose_y = attn_97_transpose_y_0, x = var_5558_cast_fp16, y = var_5556_cast_fp16)[name = tensor("attn_97_cast_fp16")]; tensor var_5561 = const()[name = tensor("op_5561"), val = tensor([1, 1280, 1, 1])]; tensor input_241_cast_fp16 = reshape(shape = var_5561, x = attn_97_cast_fp16)[name = tensor("input_241_cast_fp16")]; tensor obj_343_pad_type_0 = const()[name = tensor("obj_343_pad_type_0"), val = tensor("valid")]; tensor obj_343_strides_0 = const()[name = tensor("obj_343_strides_0"), val = tensor([1, 1])]; tensor obj_343_pad_0 = const()[name = tensor("obj_343_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_343_dilations_0 = const()[name = tensor("obj_343_dilations_0"), val = tensor([1, 1])]; tensor obj_343_groups_0 = const()[name = tensor("obj_343_groups_0"), val = tensor(1)]; tensor layers_24_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_24_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1403140352)))]; tensor layers_24_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_24_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1406417216)))]; tensor obj_343_cast_fp16 = conv(bias = layers_24_self_attn_o_proj_bias_to_fp16, dilations = obj_343_dilations_0, groups = obj_343_groups_0, pad = obj_343_pad_0, pad_type = obj_343_pad_type_0, strides = obj_343_strides_0, weight = layers_24_self_attn_o_proj_weight_to_fp16, x = input_241_cast_fp16)[name = tensor("obj_343_cast_fp16")]; tensor inputs_147_cast_fp16 = add(x = inputs_145_cast_fp16, y = obj_343_cast_fp16)[name = tensor("inputs_147_cast_fp16")]; tensor out_147_axes_0 = const()[name = tensor("out_147_axes_0"), val = tensor([1])]; tensor var_5583_to_fp16 = const()[name = tensor("op_5583_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_147_cast_fp16 = layer_norm(axes = out_147_axes_0, epsilon = var_5583_to_fp16, x = inputs_147_cast_fp16)[name = tensor("out_147_cast_fp16")]; tensor obj_345_gamma_0_to_fp16 = const()[name = tensor("obj_345_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1406419840)))]; tensor obj_345_beta_0_to_fp16 = const()[name = tensor("obj_345_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1406422464)))]; tensor obj_345_epsilon_0_to_fp16 = const()[name = tensor("obj_345_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_345_cast_fp16 = batch_norm(beta = obj_345_beta_0_to_fp16, epsilon = obj_345_epsilon_0_to_fp16, gamma = obj_345_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_147_cast_fp16)[name = tensor("obj_345_cast_fp16")]; tensor query_99_pad_type_0 = const()[name = tensor("query_99_pad_type_0"), val = tensor("valid")]; tensor query_99_strides_0 = const()[name = tensor("query_99_strides_0"), val = tensor([1, 1])]; tensor query_99_pad_0 = const()[name = tensor("query_99_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_99_dilations_0 = const()[name = tensor("query_99_dilations_0"), val = tensor([1, 1])]; tensor query_99_groups_0 = const()[name = tensor("query_99_groups_0"), val = tensor(1)]; tensor layers_24_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_24_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1406425088)))]; tensor layers_24_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_24_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1409701952)))]; tensor query_99_cast_fp16 = conv(bias = layers_24_encoder_attn_q_proj_bias_to_fp16, dilations = query_99_dilations_0, groups = query_99_groups_0, pad = query_99_pad_0, pad_type = query_99_pad_type_0, strides = query_99_strides_0, weight = layers_24_encoder_attn_q_proj_weight_to_fp16, x = obj_345_cast_fp16)[name = tensor("query_99_cast_fp16")]; tensor key_99_pad_type_0 = const()[name = tensor("key_99_pad_type_0"), val = tensor("valid")]; tensor key_99_strides_0 = const()[name = tensor("key_99_strides_0"), val = tensor([1, 1])]; tensor key_99_pad_0 = const()[name = tensor("key_99_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_99_dilations_0 = const()[name = tensor("key_99_dilations_0"), val = tensor([1, 1])]; tensor key_99_groups_0 = const()[name = tensor("key_99_groups_0"), val = tensor(1)]; tensor layers_24_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_24_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1409704576)))]; tensor key_99_cast_fp16 = conv(dilations = key_99_dilations_0, groups = key_99_groups_0, pad = key_99_pad_0, pad_type = key_99_pad_type_0, strides = key_99_strides_0, weight = layers_24_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_99_cast_fp16")]; tensor value_99_pad_type_0 = const()[name = tensor("value_99_pad_type_0"), val = tensor("valid")]; tensor value_99_strides_0 = const()[name = tensor("value_99_strides_0"), val = tensor([1, 1])]; tensor value_99_pad_0 = const()[name = tensor("value_99_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_99_dilations_0 = const()[name = tensor("value_99_dilations_0"), val = tensor([1, 1])]; tensor value_99_groups_0 = const()[name = tensor("value_99_groups_0"), val = tensor(1)]; tensor layers_24_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_24_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1412981440)))]; tensor layers_24_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_24_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1416258304)))]; tensor value_99_cast_fp16 = conv(bias = layers_24_encoder_attn_v_proj_bias_to_fp16, dilations = value_99_dilations_0, groups = value_99_groups_0, pad = value_99_pad_0, pad_type = value_99_pad_type_0, strides = value_99_strides_0, weight = layers_24_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_99_cast_fp16")]; tensor var_5619 = const()[name = tensor("op_5619"), val = tensor([1, 20, 64, 1])]; tensor mh_q_99_cast_fp16 = reshape(shape = var_5619, x = query_99_cast_fp16)[name = tensor("mh_q_99_cast_fp16")]; tensor var_5621_to_fp16 = const()[name = tensor("op_5621_to_fp16"), val = tensor(0x1p-3)]; tensor var_5622_cast_fp16 = mul(x = mh_q_99_cast_fp16, y = var_5621_to_fp16)[name = tensor("op_5622_cast_fp16")]; tensor var_5625 = const()[name = tensor("op_5625"), val = tensor([1, 20, 64, 1500])]; tensor var_5626_cast_fp16 = reshape(shape = var_5625, x = key_99_cast_fp16)[name = tensor("op_5626_cast_fp16")]; tensor mh_w_149_transpose_x_0 = const()[name = tensor("mh_w_149_transpose_x_0"), val = tensor(true)]; tensor mh_w_149_transpose_y_0 = const()[name = tensor("mh_w_149_transpose_y_0"), val = tensor(false)]; tensor mh_w_149_cast_fp16 = matmul(transpose_x = mh_w_149_transpose_x_0, transpose_y = mh_w_149_transpose_y_0, x = var_5622_cast_fp16, y = var_5626_cast_fp16)[name = tensor("mh_w_149_cast_fp16")]; tensor obj_349_cast_fp16 = softmax(axis = var_5468, x = mh_w_149_cast_fp16)[name = tensor("obj_349_cast_fp16")]; tensor var_5630 = const()[name = tensor("op_5630"), val = tensor([1, 20, 64, 1500])]; tensor var_5631_cast_fp16 = reshape(shape = var_5630, x = value_99_cast_fp16)[name = tensor("op_5631_cast_fp16")]; tensor attn_99_transpose_x_0 = const()[name = tensor("attn_99_transpose_x_0"), val = tensor(false)]; tensor attn_99_transpose_y_0 = const()[name = tensor("attn_99_transpose_y_0"), val = tensor(true)]; tensor attn_99_cast_fp16 = matmul(transpose_x = attn_99_transpose_x_0, transpose_y = attn_99_transpose_y_0, x = var_5631_cast_fp16, y = obj_349_cast_fp16)[name = tensor("attn_99_cast_fp16")]; tensor var_5634 = const()[name = tensor("op_5634"), val = tensor([1, 1280, 1, 1])]; tensor input_243_cast_fp16 = reshape(shape = var_5634, x = attn_99_cast_fp16)[name = tensor("input_243_cast_fp16")]; tensor obj_347_pad_type_0 = const()[name = tensor("obj_347_pad_type_0"), val = tensor("valid")]; tensor obj_347_strides_0 = const()[name = tensor("obj_347_strides_0"), val = tensor([1, 1])]; tensor obj_347_pad_0 = const()[name = tensor("obj_347_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_347_dilations_0 = const()[name = tensor("obj_347_dilations_0"), val = tensor([1, 1])]; tensor obj_347_groups_0 = const()[name = tensor("obj_347_groups_0"), val = tensor(1)]; tensor layers_24_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_24_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1416260928)))]; tensor layers_24_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_24_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1419537792)))]; tensor obj_347_cast_fp16 = conv(bias = layers_24_encoder_attn_o_proj_bias_to_fp16, dilations = obj_347_dilations_0, groups = obj_347_groups_0, pad = obj_347_pad_0, pad_type = obj_347_pad_type_0, strides = obj_347_strides_0, weight = layers_24_encoder_attn_o_proj_weight_to_fp16, x = input_243_cast_fp16)[name = tensor("obj_347_cast_fp16")]; tensor inputs_149_cast_fp16 = add(x = inputs_147_cast_fp16, y = obj_347_cast_fp16)[name = tensor("inputs_149_cast_fp16")]; tensor out_149_axes_0 = const()[name = tensor("out_149_axes_0"), val = tensor([1])]; tensor var_5652_to_fp16 = const()[name = tensor("op_5652_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_149_cast_fp16 = layer_norm(axes = out_149_axes_0, epsilon = var_5652_to_fp16, x = inputs_149_cast_fp16)[name = tensor("out_149_cast_fp16")]; tensor input_245_gamma_0_to_fp16 = const()[name = tensor("input_245_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1419540416)))]; tensor input_245_beta_0_to_fp16 = const()[name = tensor("input_245_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1419543040)))]; tensor input_245_epsilon_0_to_fp16 = const()[name = tensor("input_245_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_245_cast_fp16 = batch_norm(beta = input_245_beta_0_to_fp16, epsilon = input_245_epsilon_0_to_fp16, gamma = input_245_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_149_cast_fp16)[name = tensor("input_245_cast_fp16")]; tensor input_247_pad_type_0 = const()[name = tensor("input_247_pad_type_0"), val = tensor("valid")]; tensor input_247_strides_0 = const()[name = tensor("input_247_strides_0"), val = tensor([1, 1])]; tensor input_247_pad_0 = const()[name = tensor("input_247_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_247_dilations_0 = const()[name = tensor("input_247_dilations_0"), val = tensor([1, 1])]; tensor input_247_groups_0 = const()[name = tensor("input_247_groups_0"), val = tensor(1)]; tensor layers_24_fc1_weight_to_fp16 = const()[name = tensor("layers_24_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1419545664)))]; tensor layers_24_fc1_bias_to_fp16 = const()[name = tensor("layers_24_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1432652928)))]; tensor input_247_cast_fp16 = conv(bias = layers_24_fc1_bias_to_fp16, dilations = input_247_dilations_0, groups = input_247_groups_0, pad = input_247_pad_0, pad_type = input_247_pad_type_0, strides = input_247_strides_0, weight = layers_24_fc1_weight_to_fp16, x = input_245_cast_fp16)[name = tensor("input_247_cast_fp16")]; tensor input_249_mode_0 = const()[name = tensor("input_249_mode_0"), val = tensor("EXACT")]; tensor input_249_cast_fp16 = gelu(mode = input_249_mode_0, x = input_247_cast_fp16)[name = tensor("input_249_cast_fp16")]; tensor hidden_states_51_pad_type_0 = const()[name = tensor("hidden_states_51_pad_type_0"), val = tensor("valid")]; tensor hidden_states_51_strides_0 = const()[name = tensor("hidden_states_51_strides_0"), val = tensor([1, 1])]; tensor hidden_states_51_pad_0 = const()[name = tensor("hidden_states_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_51_dilations_0 = const()[name = tensor("hidden_states_51_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_51_groups_0 = const()[name = tensor("hidden_states_51_groups_0"), val = tensor(1)]; tensor layers_24_fc2_weight_to_fp16 = const()[name = tensor("layers_24_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1432663232)))]; tensor layers_24_fc2_bias_to_fp16 = const()[name = tensor("layers_24_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1445770496)))]; tensor hidden_states_51_cast_fp16 = conv(bias = layers_24_fc2_bias_to_fp16, dilations = hidden_states_51_dilations_0, groups = hidden_states_51_groups_0, pad = hidden_states_51_pad_0, pad_type = hidden_states_51_pad_type_0, strides = hidden_states_51_strides_0, weight = layers_24_fc2_weight_to_fp16, x = input_249_cast_fp16)[name = tensor("hidden_states_51_cast_fp16")]; tensor inputs_151_cast_fp16 = add(x = inputs_149_cast_fp16, y = hidden_states_51_cast_fp16)[name = tensor("inputs_151_cast_fp16")]; tensor var_5687 = const()[name = tensor("op_5687"), val = tensor(3)]; tensor out_151_axes_0 = const()[name = tensor("out_151_axes_0"), val = tensor([1])]; tensor var_5712_to_fp16 = const()[name = tensor("op_5712_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_151_cast_fp16 = layer_norm(axes = out_151_axes_0, epsilon = var_5712_to_fp16, x = inputs_151_cast_fp16)[name = tensor("out_151_cast_fp16")]; tensor obj_351_gamma_0_to_fp16 = const()[name = tensor("obj_351_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1445773120)))]; tensor obj_351_beta_0_to_fp16 = const()[name = tensor("obj_351_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1445775744)))]; tensor obj_351_epsilon_0_to_fp16 = const()[name = tensor("obj_351_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_351_cast_fp16 = batch_norm(beta = obj_351_beta_0_to_fp16, epsilon = obj_351_epsilon_0_to_fp16, gamma = obj_351_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_151_cast_fp16)[name = tensor("obj_351_cast_fp16")]; tensor query_101_pad_type_0 = const()[name = tensor("query_101_pad_type_0"), val = tensor("valid")]; tensor query_101_strides_0 = const()[name = tensor("query_101_strides_0"), val = tensor([1, 1])]; tensor query_101_pad_0 = const()[name = tensor("query_101_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_101_dilations_0 = const()[name = tensor("query_101_dilations_0"), val = tensor([1, 1])]; tensor query_101_groups_0 = const()[name = tensor("query_101_groups_0"), val = tensor(1)]; tensor layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1445778368)))]; tensor layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1449055232)))]; tensor query_101_cast_fp16 = conv(bias = layers_25_self_attn_q_proj_bias_to_fp16, dilations = query_101_dilations_0, groups = query_101_groups_0, pad = query_101_pad_0, pad_type = query_101_pad_type_0, strides = query_101_strides_0, weight = layers_25_self_attn_q_proj_weight_to_fp16, x = obj_351_cast_fp16)[name = tensor("query_101_cast_fp16")]; tensor current_key_51_pad_type_0 = const()[name = tensor("current_key_51_pad_type_0"), val = tensor("valid")]; tensor current_key_51_strides_0 = const()[name = tensor("current_key_51_strides_0"), val = tensor([1, 1])]; tensor current_key_51_pad_0 = const()[name = tensor("current_key_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_51_dilations_0 = const()[name = tensor("current_key_51_dilations_0"), val = tensor([1, 1])]; tensor current_key_51_groups_0 = const()[name = tensor("current_key_51_groups_0"), val = tensor(1)]; tensor layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1449057856)))]; tensor current_key_51_cast_fp16 = conv(dilations = current_key_51_dilations_0, groups = current_key_51_groups_0, pad = current_key_51_pad_0, pad_type = current_key_51_pad_type_0, strides = current_key_51_strides_0, weight = layers_25_self_attn_k_proj_weight_to_fp16, x = obj_351_cast_fp16)[name = tensor("current_key_51_cast_fp16")]; tensor current_value_51_pad_type_0 = const()[name = tensor("current_value_51_pad_type_0"), val = tensor("valid")]; tensor current_value_51_strides_0 = const()[name = tensor("current_value_51_strides_0"), val = tensor([1, 1])]; tensor current_value_51_pad_0 = const()[name = tensor("current_value_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_51_dilations_0 = const()[name = tensor("current_value_51_dilations_0"), val = tensor([1, 1])]; tensor current_value_51_groups_0 = const()[name = tensor("current_value_51_groups_0"), val = tensor(1)]; tensor layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1452334720)))]; tensor layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1455611584)))]; tensor current_value_51_cast_fp16 = conv(bias = layers_25_self_attn_v_proj_bias_to_fp16, dilations = current_value_51_dilations_0, groups = current_value_51_groups_0, pad = current_value_51_pad_0, pad_type = current_value_51_pad_type_0, strides = current_value_51_strides_0, weight = layers_25_self_attn_v_proj_weight_to_fp16, x = obj_351_cast_fp16)[name = tensor("current_value_51_cast_fp16")]; tensor var_5751_cast_fp16 = mul(x = var_103_cast_fp16_25, y = var_239_cast_fp16)[name = tensor("op_5751_cast_fp16")]; tensor var_5752_cast_fp16 = mul(x = current_key_51_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5752_cast_fp16")]; tensor key_101_cast_fp16 = add(x = var_5751_cast_fp16, y = var_5752_cast_fp16)[name = tensor("key_101_cast_fp16")]; tensor var_5755_cast_fp16 = mul(x = var_138_cast_fp16_25, y = var_239_cast_fp16)[name = tensor("op_5755_cast_fp16")]; tensor var_5756_cast_fp16 = mul(x = current_value_51_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5756_cast_fp16")]; tensor value_101_cast_fp16 = add(x = var_5755_cast_fp16, y = var_5756_cast_fp16)[name = tensor("value_101_cast_fp16")]; tensor var_5760 = const()[name = tensor("op_5760"), val = tensor([1, 20, 64, 1])]; tensor mh_q_101_cast_fp16 = reshape(shape = var_5760, x = query_101_cast_fp16)[name = tensor("mh_q_101_cast_fp16")]; tensor var_5762_to_fp16 = const()[name = tensor("op_5762_to_fp16"), val = tensor(0x1p-3)]; tensor var_5763_cast_fp16 = mul(x = mh_q_101_cast_fp16, y = var_5762_to_fp16)[name = tensor("op_5763_cast_fp16")]; tensor var_5766 = const()[name = tensor("op_5766"), val = tensor([1, 20, 64, 448])]; tensor var_5767_cast_fp16 = reshape(shape = var_5766, x = key_101_cast_fp16)[name = tensor("op_5767_cast_fp16")]; tensor mh_w_151_transpose_x_0 = const()[name = tensor("mh_w_151_transpose_x_0"), val = tensor(true)]; tensor mh_w_151_transpose_y_0 = const()[name = tensor("mh_w_151_transpose_y_0"), val = tensor(false)]; tensor mh_w_151_cast_fp16 = matmul(transpose_x = mh_w_151_transpose_x_0, transpose_y = mh_w_151_transpose_y_0, x = var_5763_cast_fp16, y = var_5767_cast_fp16)[name = tensor("mh_w_151_cast_fp16")]; tensor mh_w_153_cast_fp16 = add(x = mh_w_151_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_153_cast_fp16")]; tensor var_5775_cast_fp16 = softmax(axis = var_5687, x = mh_w_153_cast_fp16)[name = tensor("op_5775_cast_fp16")]; tensor var_5776 = const()[name = tensor("op_5776"), val = tensor([1, 20, 64, 448])]; tensor var_5777_cast_fp16 = reshape(shape = var_5776, x = value_101_cast_fp16)[name = tensor("op_5777_cast_fp16")]; tensor attn_101_transpose_x_0 = const()[name = tensor("attn_101_transpose_x_0"), val = tensor(false)]; tensor attn_101_transpose_y_0 = const()[name = tensor("attn_101_transpose_y_0"), val = tensor(true)]; tensor attn_101_cast_fp16 = matmul(transpose_x = attn_101_transpose_x_0, transpose_y = attn_101_transpose_y_0, x = var_5777_cast_fp16, y = var_5775_cast_fp16)[name = tensor("attn_101_cast_fp16")]; tensor var_5780 = const()[name = tensor("op_5780"), val = tensor([1, 1280, 1, 1])]; tensor input_251_cast_fp16 = reshape(shape = var_5780, x = attn_101_cast_fp16)[name = tensor("input_251_cast_fp16")]; tensor obj_357_pad_type_0 = const()[name = tensor("obj_357_pad_type_0"), val = tensor("valid")]; tensor obj_357_strides_0 = const()[name = tensor("obj_357_strides_0"), val = tensor([1, 1])]; tensor obj_357_pad_0 = const()[name = tensor("obj_357_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_357_dilations_0 = const()[name = tensor("obj_357_dilations_0"), val = tensor([1, 1])]; tensor obj_357_groups_0 = const()[name = tensor("obj_357_groups_0"), val = tensor(1)]; tensor layers_25_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_25_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1455614208)))]; tensor layers_25_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_25_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1458891072)))]; tensor obj_357_cast_fp16 = conv(bias = layers_25_self_attn_o_proj_bias_to_fp16, dilations = obj_357_dilations_0, groups = obj_357_groups_0, pad = obj_357_pad_0, pad_type = obj_357_pad_type_0, strides = obj_357_strides_0, weight = layers_25_self_attn_o_proj_weight_to_fp16, x = input_251_cast_fp16)[name = tensor("obj_357_cast_fp16")]; tensor inputs_153_cast_fp16 = add(x = inputs_151_cast_fp16, y = obj_357_cast_fp16)[name = tensor("inputs_153_cast_fp16")]; tensor out_153_axes_0 = const()[name = tensor("out_153_axes_0"), val = tensor([1])]; tensor var_5802_to_fp16 = const()[name = tensor("op_5802_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_153_cast_fp16 = layer_norm(axes = out_153_axes_0, epsilon = var_5802_to_fp16, x = inputs_153_cast_fp16)[name = tensor("out_153_cast_fp16")]; tensor obj_359_gamma_0_to_fp16 = const()[name = tensor("obj_359_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1458893696)))]; tensor obj_359_beta_0_to_fp16 = const()[name = tensor("obj_359_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1458896320)))]; tensor obj_359_epsilon_0_to_fp16 = const()[name = tensor("obj_359_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_359_cast_fp16 = batch_norm(beta = obj_359_beta_0_to_fp16, epsilon = obj_359_epsilon_0_to_fp16, gamma = obj_359_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_153_cast_fp16)[name = tensor("obj_359_cast_fp16")]; tensor query_103_pad_type_0 = const()[name = tensor("query_103_pad_type_0"), val = tensor("valid")]; tensor query_103_strides_0 = const()[name = tensor("query_103_strides_0"), val = tensor([1, 1])]; tensor query_103_pad_0 = const()[name = tensor("query_103_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_103_dilations_0 = const()[name = tensor("query_103_dilations_0"), val = tensor([1, 1])]; tensor query_103_groups_0 = const()[name = tensor("query_103_groups_0"), val = tensor(1)]; tensor layers_25_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_25_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1458898944)))]; tensor layers_25_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_25_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1462175808)))]; tensor query_103_cast_fp16 = conv(bias = layers_25_encoder_attn_q_proj_bias_to_fp16, dilations = query_103_dilations_0, groups = query_103_groups_0, pad = query_103_pad_0, pad_type = query_103_pad_type_0, strides = query_103_strides_0, weight = layers_25_encoder_attn_q_proj_weight_to_fp16, x = obj_359_cast_fp16)[name = tensor("query_103_cast_fp16")]; tensor key_103_pad_type_0 = const()[name = tensor("key_103_pad_type_0"), val = tensor("valid")]; tensor key_103_strides_0 = const()[name = tensor("key_103_strides_0"), val = tensor([1, 1])]; tensor key_103_pad_0 = const()[name = tensor("key_103_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_103_dilations_0 = const()[name = tensor("key_103_dilations_0"), val = tensor([1, 1])]; tensor key_103_groups_0 = const()[name = tensor("key_103_groups_0"), val = tensor(1)]; tensor layers_25_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_25_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1462178432)))]; tensor key_103_cast_fp16 = conv(dilations = key_103_dilations_0, groups = key_103_groups_0, pad = key_103_pad_0, pad_type = key_103_pad_type_0, strides = key_103_strides_0, weight = layers_25_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_103_cast_fp16")]; tensor value_103_pad_type_0 = const()[name = tensor("value_103_pad_type_0"), val = tensor("valid")]; tensor value_103_strides_0 = const()[name = tensor("value_103_strides_0"), val = tensor([1, 1])]; tensor value_103_pad_0 = const()[name = tensor("value_103_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_103_dilations_0 = const()[name = tensor("value_103_dilations_0"), val = tensor([1, 1])]; tensor value_103_groups_0 = const()[name = tensor("value_103_groups_0"), val = tensor(1)]; tensor layers_25_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_25_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1465455296)))]; tensor layers_25_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_25_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1468732160)))]; tensor value_103_cast_fp16 = conv(bias = layers_25_encoder_attn_v_proj_bias_to_fp16, dilations = value_103_dilations_0, groups = value_103_groups_0, pad = value_103_pad_0, pad_type = value_103_pad_type_0, strides = value_103_strides_0, weight = layers_25_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_103_cast_fp16")]; tensor var_5838 = const()[name = tensor("op_5838"), val = tensor([1, 20, 64, 1])]; tensor mh_q_103_cast_fp16 = reshape(shape = var_5838, x = query_103_cast_fp16)[name = tensor("mh_q_103_cast_fp16")]; tensor var_5840_to_fp16 = const()[name = tensor("op_5840_to_fp16"), val = tensor(0x1p-3)]; tensor var_5841_cast_fp16 = mul(x = mh_q_103_cast_fp16, y = var_5840_to_fp16)[name = tensor("op_5841_cast_fp16")]; tensor var_5844 = const()[name = tensor("op_5844"), val = tensor([1, 20, 64, 1500])]; tensor var_5845_cast_fp16 = reshape(shape = var_5844, x = key_103_cast_fp16)[name = tensor("op_5845_cast_fp16")]; tensor mh_w_155_transpose_x_0 = const()[name = tensor("mh_w_155_transpose_x_0"), val = tensor(true)]; tensor mh_w_155_transpose_y_0 = const()[name = tensor("mh_w_155_transpose_y_0"), val = tensor(false)]; tensor mh_w_155_cast_fp16 = matmul(transpose_x = mh_w_155_transpose_x_0, transpose_y = mh_w_155_transpose_y_0, x = var_5841_cast_fp16, y = var_5845_cast_fp16)[name = tensor("mh_w_155_cast_fp16")]; tensor obj_363_cast_fp16 = softmax(axis = var_5687, x = mh_w_155_cast_fp16)[name = tensor("obj_363_cast_fp16")]; tensor var_5849 = const()[name = tensor("op_5849"), val = tensor([1, 20, 64, 1500])]; tensor var_5850_cast_fp16 = reshape(shape = var_5849, x = value_103_cast_fp16)[name = tensor("op_5850_cast_fp16")]; tensor attn_103_transpose_x_0 = const()[name = tensor("attn_103_transpose_x_0"), val = tensor(false)]; tensor attn_103_transpose_y_0 = const()[name = tensor("attn_103_transpose_y_0"), val = tensor(true)]; tensor attn_103_cast_fp16 = matmul(transpose_x = attn_103_transpose_x_0, transpose_y = attn_103_transpose_y_0, x = var_5850_cast_fp16, y = obj_363_cast_fp16)[name = tensor("attn_103_cast_fp16")]; tensor var_5853 = const()[name = tensor("op_5853"), val = tensor([1, 1280, 1, 1])]; tensor input_253_cast_fp16 = reshape(shape = var_5853, x = attn_103_cast_fp16)[name = tensor("input_253_cast_fp16")]; tensor obj_361_pad_type_0 = const()[name = tensor("obj_361_pad_type_0"), val = tensor("valid")]; tensor obj_361_strides_0 = const()[name = tensor("obj_361_strides_0"), val = tensor([1, 1])]; tensor obj_361_pad_0 = const()[name = tensor("obj_361_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_361_dilations_0 = const()[name = tensor("obj_361_dilations_0"), val = tensor([1, 1])]; tensor obj_361_groups_0 = const()[name = tensor("obj_361_groups_0"), val = tensor(1)]; tensor layers_25_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_25_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1468734784)))]; tensor layers_25_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_25_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1472011648)))]; tensor obj_361_cast_fp16 = conv(bias = layers_25_encoder_attn_o_proj_bias_to_fp16, dilations = obj_361_dilations_0, groups = obj_361_groups_0, pad = obj_361_pad_0, pad_type = obj_361_pad_type_0, strides = obj_361_strides_0, weight = layers_25_encoder_attn_o_proj_weight_to_fp16, x = input_253_cast_fp16)[name = tensor("obj_361_cast_fp16")]; tensor inputs_155_cast_fp16 = add(x = inputs_153_cast_fp16, y = obj_361_cast_fp16)[name = tensor("inputs_155_cast_fp16")]; tensor out_155_axes_0 = const()[name = tensor("out_155_axes_0"), val = tensor([1])]; tensor var_5874_to_fp16 = const()[name = tensor("op_5874_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_155_cast_fp16 = layer_norm(axes = out_155_axes_0, epsilon = var_5874_to_fp16, x = inputs_155_cast_fp16)[name = tensor("out_155_cast_fp16")]; tensor input_255_gamma_0_to_fp16 = const()[name = tensor("input_255_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1472014272)))]; tensor input_255_beta_0_to_fp16 = const()[name = tensor("input_255_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1472016896)))]; tensor input_255_epsilon_0_to_fp16 = const()[name = tensor("input_255_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_255_cast_fp16 = batch_norm(beta = input_255_beta_0_to_fp16, epsilon = input_255_epsilon_0_to_fp16, gamma = input_255_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_155_cast_fp16)[name = tensor("input_255_cast_fp16")]; tensor input_257_pad_type_0 = const()[name = tensor("input_257_pad_type_0"), val = tensor("valid")]; tensor input_257_strides_0 = const()[name = tensor("input_257_strides_0"), val = tensor([1, 1])]; tensor input_257_pad_0 = const()[name = tensor("input_257_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_257_dilations_0 = const()[name = tensor("input_257_dilations_0"), val = tensor([1, 1])]; tensor input_257_groups_0 = const()[name = tensor("input_257_groups_0"), val = tensor(1)]; tensor layers_25_fc1_weight_to_fp16 = const()[name = tensor("layers_25_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1472019520)))]; tensor layers_25_fc1_bias_to_fp16 = const()[name = tensor("layers_25_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1485126784)))]; tensor input_257_cast_fp16 = conv(bias = layers_25_fc1_bias_to_fp16, dilations = input_257_dilations_0, groups = input_257_groups_0, pad = input_257_pad_0, pad_type = input_257_pad_type_0, strides = input_257_strides_0, weight = layers_25_fc1_weight_to_fp16, x = input_255_cast_fp16)[name = tensor("input_257_cast_fp16")]; tensor input_259_mode_0 = const()[name = tensor("input_259_mode_0"), val = tensor("EXACT")]; tensor input_259_cast_fp16 = gelu(mode = input_259_mode_0, x = input_257_cast_fp16)[name = tensor("input_259_cast_fp16")]; tensor hidden_states_53_pad_type_0 = const()[name = tensor("hidden_states_53_pad_type_0"), val = tensor("valid")]; tensor hidden_states_53_strides_0 = const()[name = tensor("hidden_states_53_strides_0"), val = tensor([1, 1])]; tensor hidden_states_53_pad_0 = const()[name = tensor("hidden_states_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_53_dilations_0 = const()[name = tensor("hidden_states_53_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_53_groups_0 = const()[name = tensor("hidden_states_53_groups_0"), val = tensor(1)]; tensor layers_25_fc2_weight_to_fp16 = const()[name = tensor("layers_25_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1485137088)))]; tensor layers_25_fc2_bias_to_fp16 = const()[name = tensor("layers_25_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1498244352)))]; tensor hidden_states_53_cast_fp16 = conv(bias = layers_25_fc2_bias_to_fp16, dilations = hidden_states_53_dilations_0, groups = hidden_states_53_groups_0, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = hidden_states_53_strides_0, weight = layers_25_fc2_weight_to_fp16, x = input_259_cast_fp16)[name = tensor("hidden_states_53_cast_fp16")]; tensor inputs_157_cast_fp16 = add(x = inputs_155_cast_fp16, y = hidden_states_53_cast_fp16)[name = tensor("inputs_157_cast_fp16")]; tensor var_5910 = const()[name = tensor("op_5910"), val = tensor(3)]; tensor out_157_axes_0 = const()[name = tensor("out_157_axes_0"), val = tensor([1])]; tensor var_5935_to_fp16 = const()[name = tensor("op_5935_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_157_cast_fp16 = layer_norm(axes = out_157_axes_0, epsilon = var_5935_to_fp16, x = inputs_157_cast_fp16)[name = tensor("out_157_cast_fp16")]; tensor obj_365_gamma_0_to_fp16 = const()[name = tensor("obj_365_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1498246976)))]; tensor obj_365_beta_0_to_fp16 = const()[name = tensor("obj_365_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1498249600)))]; tensor obj_365_epsilon_0_to_fp16 = const()[name = tensor("obj_365_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_365_cast_fp16 = batch_norm(beta = obj_365_beta_0_to_fp16, epsilon = obj_365_epsilon_0_to_fp16, gamma = obj_365_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_157_cast_fp16)[name = tensor("obj_365_cast_fp16")]; tensor query_105_pad_type_0 = const()[name = tensor("query_105_pad_type_0"), val = tensor("valid")]; tensor query_105_strides_0 = const()[name = tensor("query_105_strides_0"), val = tensor([1, 1])]; tensor query_105_pad_0 = const()[name = tensor("query_105_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_105_dilations_0 = const()[name = tensor("query_105_dilations_0"), val = tensor([1, 1])]; tensor query_105_groups_0 = const()[name = tensor("query_105_groups_0"), val = tensor(1)]; tensor layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1498252224)))]; tensor layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1501529088)))]; tensor query_105_cast_fp16 = conv(bias = layers_26_self_attn_q_proj_bias_to_fp16, dilations = query_105_dilations_0, groups = query_105_groups_0, pad = query_105_pad_0, pad_type = query_105_pad_type_0, strides = query_105_strides_0, weight = layers_26_self_attn_q_proj_weight_to_fp16, x = obj_365_cast_fp16)[name = tensor("query_105_cast_fp16")]; tensor current_key_53_pad_type_0 = const()[name = tensor("current_key_53_pad_type_0"), val = tensor("valid")]; tensor current_key_53_strides_0 = const()[name = tensor("current_key_53_strides_0"), val = tensor([1, 1])]; tensor current_key_53_pad_0 = const()[name = tensor("current_key_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_53_dilations_0 = const()[name = tensor("current_key_53_dilations_0"), val = tensor([1, 1])]; tensor current_key_53_groups_0 = const()[name = tensor("current_key_53_groups_0"), val = tensor(1)]; tensor layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1501531712)))]; tensor current_key_53_cast_fp16 = conv(dilations = current_key_53_dilations_0, groups = current_key_53_groups_0, pad = current_key_53_pad_0, pad_type = current_key_53_pad_type_0, strides = current_key_53_strides_0, weight = layers_26_self_attn_k_proj_weight_to_fp16, x = obj_365_cast_fp16)[name = tensor("current_key_53_cast_fp16")]; tensor current_value_53_pad_type_0 = const()[name = tensor("current_value_53_pad_type_0"), val = tensor("valid")]; tensor current_value_53_strides_0 = const()[name = tensor("current_value_53_strides_0"), val = tensor([1, 1])]; tensor current_value_53_pad_0 = const()[name = tensor("current_value_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_53_dilations_0 = const()[name = tensor("current_value_53_dilations_0"), val = tensor([1, 1])]; tensor current_value_53_groups_0 = const()[name = tensor("current_value_53_groups_0"), val = tensor(1)]; tensor layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1504808576)))]; tensor layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1508085440)))]; tensor current_value_53_cast_fp16 = conv(bias = layers_26_self_attn_v_proj_bias_to_fp16, dilations = current_value_53_dilations_0, groups = current_value_53_groups_0, pad = current_value_53_pad_0, pad_type = current_value_53_pad_type_0, strides = current_value_53_strides_0, weight = layers_26_self_attn_v_proj_weight_to_fp16, x = obj_365_cast_fp16)[name = tensor("current_value_53_cast_fp16")]; tensor var_5974_cast_fp16 = mul(x = var_103_cast_fp16_26, y = var_239_cast_fp16)[name = tensor("op_5974_cast_fp16")]; tensor var_5975_cast_fp16 = mul(x = current_key_53_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5975_cast_fp16")]; tensor key_105_cast_fp16 = add(x = var_5974_cast_fp16, y = var_5975_cast_fp16)[name = tensor("key_105_cast_fp16")]; tensor var_5978_cast_fp16 = mul(x = var_138_cast_fp16_26, y = var_239_cast_fp16)[name = tensor("op_5978_cast_fp16")]; tensor var_5979_cast_fp16 = mul(x = current_value_53_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_5979_cast_fp16")]; tensor value_105_cast_fp16 = add(x = var_5978_cast_fp16, y = var_5979_cast_fp16)[name = tensor("value_105_cast_fp16")]; tensor var_5983 = const()[name = tensor("op_5983"), val = tensor([1, 20, 64, 1])]; tensor mh_q_105_cast_fp16 = reshape(shape = var_5983, x = query_105_cast_fp16)[name = tensor("mh_q_105_cast_fp16")]; tensor var_5985_to_fp16 = const()[name = tensor("op_5985_to_fp16"), val = tensor(0x1p-3)]; tensor var_5986_cast_fp16 = mul(x = mh_q_105_cast_fp16, y = var_5985_to_fp16)[name = tensor("op_5986_cast_fp16")]; tensor var_5989 = const()[name = tensor("op_5989"), val = tensor([1, 20, 64, 448])]; tensor var_5990_cast_fp16 = reshape(shape = var_5989, x = key_105_cast_fp16)[name = tensor("op_5990_cast_fp16")]; tensor mh_w_157_transpose_x_0 = const()[name = tensor("mh_w_157_transpose_x_0"), val = tensor(true)]; tensor mh_w_157_transpose_y_0 = const()[name = tensor("mh_w_157_transpose_y_0"), val = tensor(false)]; tensor mh_w_157_cast_fp16 = matmul(transpose_x = mh_w_157_transpose_x_0, transpose_y = mh_w_157_transpose_y_0, x = var_5986_cast_fp16, y = var_5990_cast_fp16)[name = tensor("mh_w_157_cast_fp16")]; tensor mh_w_159_cast_fp16 = add(x = mh_w_157_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_159_cast_fp16")]; tensor var_5998_cast_fp16 = softmax(axis = var_5910, x = mh_w_159_cast_fp16)[name = tensor("op_5998_cast_fp16")]; tensor var_5999 = const()[name = tensor("op_5999"), val = tensor([1, 20, 64, 448])]; tensor var_6000_cast_fp16 = reshape(shape = var_5999, x = value_105_cast_fp16)[name = tensor("op_6000_cast_fp16")]; tensor attn_105_transpose_x_0 = const()[name = tensor("attn_105_transpose_x_0"), val = tensor(false)]; tensor attn_105_transpose_y_0 = const()[name = tensor("attn_105_transpose_y_0"), val = tensor(true)]; tensor attn_105_cast_fp16 = matmul(transpose_x = attn_105_transpose_x_0, transpose_y = attn_105_transpose_y_0, x = var_6000_cast_fp16, y = var_5998_cast_fp16)[name = tensor("attn_105_cast_fp16")]; tensor var_6003 = const()[name = tensor("op_6003"), val = tensor([1, 1280, 1, 1])]; tensor input_261_cast_fp16 = reshape(shape = var_6003, x = attn_105_cast_fp16)[name = tensor("input_261_cast_fp16")]; tensor obj_371_pad_type_0 = const()[name = tensor("obj_371_pad_type_0"), val = tensor("valid")]; tensor obj_371_strides_0 = const()[name = tensor("obj_371_strides_0"), val = tensor([1, 1])]; tensor obj_371_pad_0 = const()[name = tensor("obj_371_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_371_dilations_0 = const()[name = tensor("obj_371_dilations_0"), val = tensor([1, 1])]; tensor obj_371_groups_0 = const()[name = tensor("obj_371_groups_0"), val = tensor(1)]; tensor layers_26_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_26_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1508088064)))]; tensor layers_26_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_26_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1511364928)))]; tensor obj_371_cast_fp16 = conv(bias = layers_26_self_attn_o_proj_bias_to_fp16, dilations = obj_371_dilations_0, groups = obj_371_groups_0, pad = obj_371_pad_0, pad_type = obj_371_pad_type_0, strides = obj_371_strides_0, weight = layers_26_self_attn_o_proj_weight_to_fp16, x = input_261_cast_fp16)[name = tensor("obj_371_cast_fp16")]; tensor inputs_159_cast_fp16 = add(x = inputs_157_cast_fp16, y = obj_371_cast_fp16)[name = tensor("inputs_159_cast_fp16")]; tensor out_159_axes_0 = const()[name = tensor("out_159_axes_0"), val = tensor([1])]; tensor var_6025_to_fp16 = const()[name = tensor("op_6025_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_159_cast_fp16 = layer_norm(axes = out_159_axes_0, epsilon = var_6025_to_fp16, x = inputs_159_cast_fp16)[name = tensor("out_159_cast_fp16")]; tensor obj_373_gamma_0_to_fp16 = const()[name = tensor("obj_373_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1511367552)))]; tensor obj_373_beta_0_to_fp16 = const()[name = tensor("obj_373_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1511370176)))]; tensor obj_373_epsilon_0_to_fp16 = const()[name = tensor("obj_373_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_373_cast_fp16 = batch_norm(beta = obj_373_beta_0_to_fp16, epsilon = obj_373_epsilon_0_to_fp16, gamma = obj_373_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_159_cast_fp16)[name = tensor("obj_373_cast_fp16")]; tensor query_107_pad_type_0 = const()[name = tensor("query_107_pad_type_0"), val = tensor("valid")]; tensor query_107_strides_0 = const()[name = tensor("query_107_strides_0"), val = tensor([1, 1])]; tensor query_107_pad_0 = const()[name = tensor("query_107_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_107_dilations_0 = const()[name = tensor("query_107_dilations_0"), val = tensor([1, 1])]; tensor query_107_groups_0 = const()[name = tensor("query_107_groups_0"), val = tensor(1)]; tensor layers_26_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_26_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1511372800)))]; tensor layers_26_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_26_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1514649664)))]; tensor query_107_cast_fp16 = conv(bias = layers_26_encoder_attn_q_proj_bias_to_fp16, dilations = query_107_dilations_0, groups = query_107_groups_0, pad = query_107_pad_0, pad_type = query_107_pad_type_0, strides = query_107_strides_0, weight = layers_26_encoder_attn_q_proj_weight_to_fp16, x = obj_373_cast_fp16)[name = tensor("query_107_cast_fp16")]; tensor key_107_pad_type_0 = const()[name = tensor("key_107_pad_type_0"), val = tensor("valid")]; tensor key_107_strides_0 = const()[name = tensor("key_107_strides_0"), val = tensor([1, 1])]; tensor key_107_pad_0 = const()[name = tensor("key_107_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_107_dilations_0 = const()[name = tensor("key_107_dilations_0"), val = tensor([1, 1])]; tensor key_107_groups_0 = const()[name = tensor("key_107_groups_0"), val = tensor(1)]; tensor layers_26_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_26_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1514652288)))]; tensor key_107_cast_fp16 = conv(dilations = key_107_dilations_0, groups = key_107_groups_0, pad = key_107_pad_0, pad_type = key_107_pad_type_0, strides = key_107_strides_0, weight = layers_26_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_107_cast_fp16")]; tensor value_107_pad_type_0 = const()[name = tensor("value_107_pad_type_0"), val = tensor("valid")]; tensor value_107_strides_0 = const()[name = tensor("value_107_strides_0"), val = tensor([1, 1])]; tensor value_107_pad_0 = const()[name = tensor("value_107_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_107_dilations_0 = const()[name = tensor("value_107_dilations_0"), val = tensor([1, 1])]; tensor value_107_groups_0 = const()[name = tensor("value_107_groups_0"), val = tensor(1)]; tensor layers_26_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_26_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1517929152)))]; tensor layers_26_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_26_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1521206016)))]; tensor value_107_cast_fp16 = conv(bias = layers_26_encoder_attn_v_proj_bias_to_fp16, dilations = value_107_dilations_0, groups = value_107_groups_0, pad = value_107_pad_0, pad_type = value_107_pad_type_0, strides = value_107_strides_0, weight = layers_26_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_107_cast_fp16")]; tensor var_6061 = const()[name = tensor("op_6061"), val = tensor([1, 20, 64, 1])]; tensor mh_q_107_cast_fp16 = reshape(shape = var_6061, x = query_107_cast_fp16)[name = tensor("mh_q_107_cast_fp16")]; tensor var_6063_to_fp16 = const()[name = tensor("op_6063_to_fp16"), val = tensor(0x1p-3)]; tensor var_6064_cast_fp16 = mul(x = mh_q_107_cast_fp16, y = var_6063_to_fp16)[name = tensor("op_6064_cast_fp16")]; tensor var_6067 = const()[name = tensor("op_6067"), val = tensor([1, 20, 64, 1500])]; tensor var_6068_cast_fp16 = reshape(shape = var_6067, x = key_107_cast_fp16)[name = tensor("op_6068_cast_fp16")]; tensor mh_w_161_transpose_x_0 = const()[name = tensor("mh_w_161_transpose_x_0"), val = tensor(true)]; tensor mh_w_161_transpose_y_0 = const()[name = tensor("mh_w_161_transpose_y_0"), val = tensor(false)]; tensor mh_w_161_cast_fp16 = matmul(transpose_x = mh_w_161_transpose_x_0, transpose_y = mh_w_161_transpose_y_0, x = var_6064_cast_fp16, y = var_6068_cast_fp16)[name = tensor("mh_w_161_cast_fp16")]; tensor obj_377_cast_fp16 = softmax(axis = var_5910, x = mh_w_161_cast_fp16)[name = tensor("obj_377_cast_fp16")]; tensor var_6072 = const()[name = tensor("op_6072"), val = tensor([1, 20, 64, 1500])]; tensor var_6073_cast_fp16 = reshape(shape = var_6072, x = value_107_cast_fp16)[name = tensor("op_6073_cast_fp16")]; tensor attn_107_transpose_x_0 = const()[name = tensor("attn_107_transpose_x_0"), val = tensor(false)]; tensor attn_107_transpose_y_0 = const()[name = tensor("attn_107_transpose_y_0"), val = tensor(true)]; tensor attn_107_cast_fp16 = matmul(transpose_x = attn_107_transpose_x_0, transpose_y = attn_107_transpose_y_0, x = var_6073_cast_fp16, y = obj_377_cast_fp16)[name = tensor("attn_107_cast_fp16")]; tensor var_6076 = const()[name = tensor("op_6076"), val = tensor([1, 1280, 1, 1])]; tensor input_263_cast_fp16 = reshape(shape = var_6076, x = attn_107_cast_fp16)[name = tensor("input_263_cast_fp16")]; tensor obj_375_pad_type_0 = const()[name = tensor("obj_375_pad_type_0"), val = tensor("valid")]; tensor obj_375_strides_0 = const()[name = tensor("obj_375_strides_0"), val = tensor([1, 1])]; tensor obj_375_pad_0 = const()[name = tensor("obj_375_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_375_dilations_0 = const()[name = tensor("obj_375_dilations_0"), val = tensor([1, 1])]; tensor obj_375_groups_0 = const()[name = tensor("obj_375_groups_0"), val = tensor(1)]; tensor layers_26_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_26_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1521208640)))]; tensor layers_26_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_26_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1524485504)))]; tensor obj_375_cast_fp16 = conv(bias = layers_26_encoder_attn_o_proj_bias_to_fp16, dilations = obj_375_dilations_0, groups = obj_375_groups_0, pad = obj_375_pad_0, pad_type = obj_375_pad_type_0, strides = obj_375_strides_0, weight = layers_26_encoder_attn_o_proj_weight_to_fp16, x = input_263_cast_fp16)[name = tensor("obj_375_cast_fp16")]; tensor inputs_161_cast_fp16 = add(x = inputs_159_cast_fp16, y = obj_375_cast_fp16)[name = tensor("inputs_161_cast_fp16")]; tensor out_161_axes_0 = const()[name = tensor("out_161_axes_0"), val = tensor([1])]; tensor var_6097_to_fp16 = const()[name = tensor("op_6097_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_161_cast_fp16 = layer_norm(axes = out_161_axes_0, epsilon = var_6097_to_fp16, x = inputs_161_cast_fp16)[name = tensor("out_161_cast_fp16")]; tensor input_265_gamma_0_to_fp16 = const()[name = tensor("input_265_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1524488128)))]; tensor input_265_beta_0_to_fp16 = const()[name = tensor("input_265_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1524490752)))]; tensor input_265_epsilon_0_to_fp16 = const()[name = tensor("input_265_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_265_cast_fp16 = batch_norm(beta = input_265_beta_0_to_fp16, epsilon = input_265_epsilon_0_to_fp16, gamma = input_265_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_161_cast_fp16)[name = tensor("input_265_cast_fp16")]; tensor input_267_pad_type_0 = const()[name = tensor("input_267_pad_type_0"), val = tensor("valid")]; tensor input_267_strides_0 = const()[name = tensor("input_267_strides_0"), val = tensor([1, 1])]; tensor input_267_pad_0 = const()[name = tensor("input_267_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_267_dilations_0 = const()[name = tensor("input_267_dilations_0"), val = tensor([1, 1])]; tensor input_267_groups_0 = const()[name = tensor("input_267_groups_0"), val = tensor(1)]; tensor layers_26_fc1_weight_to_fp16 = const()[name = tensor("layers_26_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1524493376)))]; tensor layers_26_fc1_bias_to_fp16 = const()[name = tensor("layers_26_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1537600640)))]; tensor input_267_cast_fp16 = conv(bias = layers_26_fc1_bias_to_fp16, dilations = input_267_dilations_0, groups = input_267_groups_0, pad = input_267_pad_0, pad_type = input_267_pad_type_0, strides = input_267_strides_0, weight = layers_26_fc1_weight_to_fp16, x = input_265_cast_fp16)[name = tensor("input_267_cast_fp16")]; tensor input_269_mode_0 = const()[name = tensor("input_269_mode_0"), val = tensor("EXACT")]; tensor input_269_cast_fp16 = gelu(mode = input_269_mode_0, x = input_267_cast_fp16)[name = tensor("input_269_cast_fp16")]; tensor hidden_states_55_pad_type_0 = const()[name = tensor("hidden_states_55_pad_type_0"), val = tensor("valid")]; tensor hidden_states_55_strides_0 = const()[name = tensor("hidden_states_55_strides_0"), val = tensor([1, 1])]; tensor hidden_states_55_pad_0 = const()[name = tensor("hidden_states_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_55_dilations_0 = const()[name = tensor("hidden_states_55_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_55_groups_0 = const()[name = tensor("hidden_states_55_groups_0"), val = tensor(1)]; tensor layers_26_fc2_weight_to_fp16 = const()[name = tensor("layers_26_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1537610944)))]; tensor layers_26_fc2_bias_to_fp16 = const()[name = tensor("layers_26_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1550718208)))]; tensor hidden_states_55_cast_fp16 = conv(bias = layers_26_fc2_bias_to_fp16, dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = layers_26_fc2_weight_to_fp16, x = input_269_cast_fp16)[name = tensor("hidden_states_55_cast_fp16")]; tensor inputs_163_cast_fp16 = add(x = inputs_161_cast_fp16, y = hidden_states_55_cast_fp16)[name = tensor("inputs_163_cast_fp16")]; tensor var_6133 = const()[name = tensor("op_6133"), val = tensor(3)]; tensor out_163_axes_0 = const()[name = tensor("out_163_axes_0"), val = tensor([1])]; tensor var_6158_to_fp16 = const()[name = tensor("op_6158_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_163_cast_fp16 = layer_norm(axes = out_163_axes_0, epsilon = var_6158_to_fp16, x = inputs_163_cast_fp16)[name = tensor("out_163_cast_fp16")]; tensor obj_379_gamma_0_to_fp16 = const()[name = tensor("obj_379_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1550720832)))]; tensor obj_379_beta_0_to_fp16 = const()[name = tensor("obj_379_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1550723456)))]; tensor obj_379_epsilon_0_to_fp16 = const()[name = tensor("obj_379_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_379_cast_fp16 = batch_norm(beta = obj_379_beta_0_to_fp16, epsilon = obj_379_epsilon_0_to_fp16, gamma = obj_379_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_163_cast_fp16)[name = tensor("obj_379_cast_fp16")]; tensor query_109_pad_type_0 = const()[name = tensor("query_109_pad_type_0"), val = tensor("valid")]; tensor query_109_strides_0 = const()[name = tensor("query_109_strides_0"), val = tensor([1, 1])]; tensor query_109_pad_0 = const()[name = tensor("query_109_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_109_dilations_0 = const()[name = tensor("query_109_dilations_0"), val = tensor([1, 1])]; tensor query_109_groups_0 = const()[name = tensor("query_109_groups_0"), val = tensor(1)]; tensor layers_27_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_27_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1550726080)))]; tensor layers_27_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_27_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1554002944)))]; tensor query_109_cast_fp16 = conv(bias = layers_27_self_attn_q_proj_bias_to_fp16, dilations = query_109_dilations_0, groups = query_109_groups_0, pad = query_109_pad_0, pad_type = query_109_pad_type_0, strides = query_109_strides_0, weight = layers_27_self_attn_q_proj_weight_to_fp16, x = obj_379_cast_fp16)[name = tensor("query_109_cast_fp16")]; tensor current_key_55_pad_type_0 = const()[name = tensor("current_key_55_pad_type_0"), val = tensor("valid")]; tensor current_key_55_strides_0 = const()[name = tensor("current_key_55_strides_0"), val = tensor([1, 1])]; tensor current_key_55_pad_0 = const()[name = tensor("current_key_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_55_dilations_0 = const()[name = tensor("current_key_55_dilations_0"), val = tensor([1, 1])]; tensor current_key_55_groups_0 = const()[name = tensor("current_key_55_groups_0"), val = tensor(1)]; tensor layers_27_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_27_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1554005568)))]; tensor current_key_55_cast_fp16 = conv(dilations = current_key_55_dilations_0, groups = current_key_55_groups_0, pad = current_key_55_pad_0, pad_type = current_key_55_pad_type_0, strides = current_key_55_strides_0, weight = layers_27_self_attn_k_proj_weight_to_fp16, x = obj_379_cast_fp16)[name = tensor("current_key_55_cast_fp16")]; tensor current_value_55_pad_type_0 = const()[name = tensor("current_value_55_pad_type_0"), val = tensor("valid")]; tensor current_value_55_strides_0 = const()[name = tensor("current_value_55_strides_0"), val = tensor([1, 1])]; tensor current_value_55_pad_0 = const()[name = tensor("current_value_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_55_dilations_0 = const()[name = tensor("current_value_55_dilations_0"), val = tensor([1, 1])]; tensor current_value_55_groups_0 = const()[name = tensor("current_value_55_groups_0"), val = tensor(1)]; tensor layers_27_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_27_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1557282432)))]; tensor layers_27_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_27_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1560559296)))]; tensor current_value_55_cast_fp16 = conv(bias = layers_27_self_attn_v_proj_bias_to_fp16, dilations = current_value_55_dilations_0, groups = current_value_55_groups_0, pad = current_value_55_pad_0, pad_type = current_value_55_pad_type_0, strides = current_value_55_strides_0, weight = layers_27_self_attn_v_proj_weight_to_fp16, x = obj_379_cast_fp16)[name = tensor("current_value_55_cast_fp16")]; tensor var_6197_cast_fp16 = mul(x = var_103_cast_fp16_27, y = var_239_cast_fp16)[name = tensor("op_6197_cast_fp16")]; tensor var_6198_cast_fp16 = mul(x = current_key_55_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_6198_cast_fp16")]; tensor key_109_cast_fp16 = add(x = var_6197_cast_fp16, y = var_6198_cast_fp16)[name = tensor("key_109_cast_fp16")]; tensor var_6201_cast_fp16 = mul(x = var_138_cast_fp16_27, y = var_239_cast_fp16)[name = tensor("op_6201_cast_fp16")]; tensor var_6202_cast_fp16 = mul(x = current_value_55_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_6202_cast_fp16")]; tensor value_109_cast_fp16 = add(x = var_6201_cast_fp16, y = var_6202_cast_fp16)[name = tensor("value_109_cast_fp16")]; tensor var_6206 = const()[name = tensor("op_6206"), val = tensor([1, 20, 64, 1])]; tensor mh_q_109_cast_fp16 = reshape(shape = var_6206, x = query_109_cast_fp16)[name = tensor("mh_q_109_cast_fp16")]; tensor var_6208_to_fp16 = const()[name = tensor("op_6208_to_fp16"), val = tensor(0x1p-3)]; tensor var_6209_cast_fp16 = mul(x = mh_q_109_cast_fp16, y = var_6208_to_fp16)[name = tensor("op_6209_cast_fp16")]; tensor var_6212 = const()[name = tensor("op_6212"), val = tensor([1, 20, 64, 448])]; tensor var_6213_cast_fp16 = reshape(shape = var_6212, x = key_109_cast_fp16)[name = tensor("op_6213_cast_fp16")]; tensor mh_w_163_transpose_x_0 = const()[name = tensor("mh_w_163_transpose_x_0"), val = tensor(true)]; tensor mh_w_163_transpose_y_0 = const()[name = tensor("mh_w_163_transpose_y_0"), val = tensor(false)]; tensor mh_w_163_cast_fp16 = matmul(transpose_x = mh_w_163_transpose_x_0, transpose_y = mh_w_163_transpose_y_0, x = var_6209_cast_fp16, y = var_6213_cast_fp16)[name = tensor("mh_w_163_cast_fp16")]; tensor mh_w_165_cast_fp16 = add(x = mh_w_163_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_165_cast_fp16")]; tensor var_6221_cast_fp16 = softmax(axis = var_6133, x = mh_w_165_cast_fp16)[name = tensor("op_6221_cast_fp16")]; tensor var_6222 = const()[name = tensor("op_6222"), val = tensor([1, 20, 64, 448])]; tensor var_6223_cast_fp16 = reshape(shape = var_6222, x = value_109_cast_fp16)[name = tensor("op_6223_cast_fp16")]; tensor attn_109_transpose_x_0 = const()[name = tensor("attn_109_transpose_x_0"), val = tensor(false)]; tensor attn_109_transpose_y_0 = const()[name = tensor("attn_109_transpose_y_0"), val = tensor(true)]; tensor attn_109_cast_fp16 = matmul(transpose_x = attn_109_transpose_x_0, transpose_y = attn_109_transpose_y_0, x = var_6223_cast_fp16, y = var_6221_cast_fp16)[name = tensor("attn_109_cast_fp16")]; tensor var_6226 = const()[name = tensor("op_6226"), val = tensor([1, 1280, 1, 1])]; tensor input_271_cast_fp16 = reshape(shape = var_6226, x = attn_109_cast_fp16)[name = tensor("input_271_cast_fp16")]; tensor obj_385_pad_type_0 = const()[name = tensor("obj_385_pad_type_0"), val = tensor("valid")]; tensor obj_385_strides_0 = const()[name = tensor("obj_385_strides_0"), val = tensor([1, 1])]; tensor obj_385_pad_0 = const()[name = tensor("obj_385_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_385_dilations_0 = const()[name = tensor("obj_385_dilations_0"), val = tensor([1, 1])]; tensor obj_385_groups_0 = const()[name = tensor("obj_385_groups_0"), val = tensor(1)]; tensor layers_27_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_27_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1560561920)))]; tensor layers_27_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_27_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1563838784)))]; tensor obj_385_cast_fp16 = conv(bias = layers_27_self_attn_o_proj_bias_to_fp16, dilations = obj_385_dilations_0, groups = obj_385_groups_0, pad = obj_385_pad_0, pad_type = obj_385_pad_type_0, strides = obj_385_strides_0, weight = layers_27_self_attn_o_proj_weight_to_fp16, x = input_271_cast_fp16)[name = tensor("obj_385_cast_fp16")]; tensor inputs_165_cast_fp16 = add(x = inputs_163_cast_fp16, y = obj_385_cast_fp16)[name = tensor("inputs_165_cast_fp16")]; tensor out_165_axes_0 = const()[name = tensor("out_165_axes_0"), val = tensor([1])]; tensor var_6248_to_fp16 = const()[name = tensor("op_6248_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_165_cast_fp16 = layer_norm(axes = out_165_axes_0, epsilon = var_6248_to_fp16, x = inputs_165_cast_fp16)[name = tensor("out_165_cast_fp16")]; tensor obj_387_gamma_0_to_fp16 = const()[name = tensor("obj_387_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1563841408)))]; tensor obj_387_beta_0_to_fp16 = const()[name = tensor("obj_387_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1563844032)))]; tensor obj_387_epsilon_0_to_fp16 = const()[name = tensor("obj_387_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_387_cast_fp16 = batch_norm(beta = obj_387_beta_0_to_fp16, epsilon = obj_387_epsilon_0_to_fp16, gamma = obj_387_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_165_cast_fp16)[name = tensor("obj_387_cast_fp16")]; tensor query_111_pad_type_0 = const()[name = tensor("query_111_pad_type_0"), val = tensor("valid")]; tensor query_111_strides_0 = const()[name = tensor("query_111_strides_0"), val = tensor([1, 1])]; tensor query_111_pad_0 = const()[name = tensor("query_111_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_111_dilations_0 = const()[name = tensor("query_111_dilations_0"), val = tensor([1, 1])]; tensor query_111_groups_0 = const()[name = tensor("query_111_groups_0"), val = tensor(1)]; tensor layers_27_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_27_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1563846656)))]; tensor layers_27_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_27_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1567123520)))]; tensor query_111_cast_fp16 = conv(bias = layers_27_encoder_attn_q_proj_bias_to_fp16, dilations = query_111_dilations_0, groups = query_111_groups_0, pad = query_111_pad_0, pad_type = query_111_pad_type_0, strides = query_111_strides_0, weight = layers_27_encoder_attn_q_proj_weight_to_fp16, x = obj_387_cast_fp16)[name = tensor("query_111_cast_fp16")]; tensor key_111_pad_type_0 = const()[name = tensor("key_111_pad_type_0"), val = tensor("valid")]; tensor key_111_strides_0 = const()[name = tensor("key_111_strides_0"), val = tensor([1, 1])]; tensor key_111_pad_0 = const()[name = tensor("key_111_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_111_dilations_0 = const()[name = tensor("key_111_dilations_0"), val = tensor([1, 1])]; tensor key_111_groups_0 = const()[name = tensor("key_111_groups_0"), val = tensor(1)]; tensor layers_27_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_27_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1567126144)))]; tensor key_111_cast_fp16 = conv(dilations = key_111_dilations_0, groups = key_111_groups_0, pad = key_111_pad_0, pad_type = key_111_pad_type_0, strides = key_111_strides_0, weight = layers_27_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_111_cast_fp16")]; tensor value_111_pad_type_0 = const()[name = tensor("value_111_pad_type_0"), val = tensor("valid")]; tensor value_111_strides_0 = const()[name = tensor("value_111_strides_0"), val = tensor([1, 1])]; tensor value_111_pad_0 = const()[name = tensor("value_111_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_111_dilations_0 = const()[name = tensor("value_111_dilations_0"), val = tensor([1, 1])]; tensor value_111_groups_0 = const()[name = tensor("value_111_groups_0"), val = tensor(1)]; tensor layers_27_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_27_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1570403008)))]; tensor layers_27_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_27_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1573679872)))]; tensor value_111_cast_fp16 = conv(bias = layers_27_encoder_attn_v_proj_bias_to_fp16, dilations = value_111_dilations_0, groups = value_111_groups_0, pad = value_111_pad_0, pad_type = value_111_pad_type_0, strides = value_111_strides_0, weight = layers_27_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_111_cast_fp16")]; tensor var_6284 = const()[name = tensor("op_6284"), val = tensor([1, 20, 64, 1])]; tensor mh_q_111_cast_fp16 = reshape(shape = var_6284, x = query_111_cast_fp16)[name = tensor("mh_q_111_cast_fp16")]; tensor var_6286_to_fp16 = const()[name = tensor("op_6286_to_fp16"), val = tensor(0x1p-3)]; tensor var_6287_cast_fp16 = mul(x = mh_q_111_cast_fp16, y = var_6286_to_fp16)[name = tensor("op_6287_cast_fp16")]; tensor var_6290 = const()[name = tensor("op_6290"), val = tensor([1, 20, 64, 1500])]; tensor var_6291_cast_fp16 = reshape(shape = var_6290, x = key_111_cast_fp16)[name = tensor("op_6291_cast_fp16")]; tensor mh_w_167_transpose_x_0 = const()[name = tensor("mh_w_167_transpose_x_0"), val = tensor(true)]; tensor mh_w_167_transpose_y_0 = const()[name = tensor("mh_w_167_transpose_y_0"), val = tensor(false)]; tensor mh_w_167_cast_fp16 = matmul(transpose_x = mh_w_167_transpose_x_0, transpose_y = mh_w_167_transpose_y_0, x = var_6287_cast_fp16, y = var_6291_cast_fp16)[name = tensor("mh_w_167_cast_fp16")]; tensor obj_391_cast_fp16 = softmax(axis = var_6133, x = mh_w_167_cast_fp16)[name = tensor("obj_391_cast_fp16")]; tensor var_6295 = const()[name = tensor("op_6295"), val = tensor([1, 20, 64, 1500])]; tensor var_6296_cast_fp16 = reshape(shape = var_6295, x = value_111_cast_fp16)[name = tensor("op_6296_cast_fp16")]; tensor attn_111_transpose_x_0 = const()[name = tensor("attn_111_transpose_x_0"), val = tensor(false)]; tensor attn_111_transpose_y_0 = const()[name = tensor("attn_111_transpose_y_0"), val = tensor(true)]; tensor attn_111_cast_fp16 = matmul(transpose_x = attn_111_transpose_x_0, transpose_y = attn_111_transpose_y_0, x = var_6296_cast_fp16, y = obj_391_cast_fp16)[name = tensor("attn_111_cast_fp16")]; tensor var_6299 = const()[name = tensor("op_6299"), val = tensor([1, 1280, 1, 1])]; tensor input_273_cast_fp16 = reshape(shape = var_6299, x = attn_111_cast_fp16)[name = tensor("input_273_cast_fp16")]; tensor obj_389_pad_type_0 = const()[name = tensor("obj_389_pad_type_0"), val = tensor("valid")]; tensor obj_389_strides_0 = const()[name = tensor("obj_389_strides_0"), val = tensor([1, 1])]; tensor obj_389_pad_0 = const()[name = tensor("obj_389_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_389_dilations_0 = const()[name = tensor("obj_389_dilations_0"), val = tensor([1, 1])]; tensor obj_389_groups_0 = const()[name = tensor("obj_389_groups_0"), val = tensor(1)]; tensor layers_27_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_27_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1573682496)))]; tensor layers_27_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_27_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1576959360)))]; tensor obj_389_cast_fp16 = conv(bias = layers_27_encoder_attn_o_proj_bias_to_fp16, dilations = obj_389_dilations_0, groups = obj_389_groups_0, pad = obj_389_pad_0, pad_type = obj_389_pad_type_0, strides = obj_389_strides_0, weight = layers_27_encoder_attn_o_proj_weight_to_fp16, x = input_273_cast_fp16)[name = tensor("obj_389_cast_fp16")]; tensor inputs_167_cast_fp16 = add(x = inputs_165_cast_fp16, y = obj_389_cast_fp16)[name = tensor("inputs_167_cast_fp16")]; tensor out_167_axes_0 = const()[name = tensor("out_167_axes_0"), val = tensor([1])]; tensor var_6320_to_fp16 = const()[name = tensor("op_6320_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_167_cast_fp16 = layer_norm(axes = out_167_axes_0, epsilon = var_6320_to_fp16, x = inputs_167_cast_fp16)[name = tensor("out_167_cast_fp16")]; tensor input_275_gamma_0_to_fp16 = const()[name = tensor("input_275_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1576961984)))]; tensor input_275_beta_0_to_fp16 = const()[name = tensor("input_275_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1576964608)))]; tensor input_275_epsilon_0_to_fp16 = const()[name = tensor("input_275_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_275_cast_fp16 = batch_norm(beta = input_275_beta_0_to_fp16, epsilon = input_275_epsilon_0_to_fp16, gamma = input_275_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_167_cast_fp16)[name = tensor("input_275_cast_fp16")]; tensor input_277_pad_type_0 = const()[name = tensor("input_277_pad_type_0"), val = tensor("valid")]; tensor input_277_strides_0 = const()[name = tensor("input_277_strides_0"), val = tensor([1, 1])]; tensor input_277_pad_0 = const()[name = tensor("input_277_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_277_dilations_0 = const()[name = tensor("input_277_dilations_0"), val = tensor([1, 1])]; tensor input_277_groups_0 = const()[name = tensor("input_277_groups_0"), val = tensor(1)]; tensor layers_27_fc1_weight_to_fp16 = const()[name = tensor("layers_27_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1576967232)))]; tensor layers_27_fc1_bias_to_fp16 = const()[name = tensor("layers_27_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1590074496)))]; tensor input_277_cast_fp16 = conv(bias = layers_27_fc1_bias_to_fp16, dilations = input_277_dilations_0, groups = input_277_groups_0, pad = input_277_pad_0, pad_type = input_277_pad_type_0, strides = input_277_strides_0, weight = layers_27_fc1_weight_to_fp16, x = input_275_cast_fp16)[name = tensor("input_277_cast_fp16")]; tensor input_279_mode_0 = const()[name = tensor("input_279_mode_0"), val = tensor("EXACT")]; tensor input_279_cast_fp16 = gelu(mode = input_279_mode_0, x = input_277_cast_fp16)[name = tensor("input_279_cast_fp16")]; tensor hidden_states_57_pad_type_0 = const()[name = tensor("hidden_states_57_pad_type_0"), val = tensor("valid")]; tensor hidden_states_57_strides_0 = const()[name = tensor("hidden_states_57_strides_0"), val = tensor([1, 1])]; tensor hidden_states_57_pad_0 = const()[name = tensor("hidden_states_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_57_dilations_0 = const()[name = tensor("hidden_states_57_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_57_groups_0 = const()[name = tensor("hidden_states_57_groups_0"), val = tensor(1)]; tensor layers_27_fc2_weight_to_fp16 = const()[name = tensor("layers_27_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1590084800)))]; tensor layers_27_fc2_bias_to_fp16 = const()[name = tensor("layers_27_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1603192064)))]; tensor hidden_states_57_cast_fp16 = conv(bias = layers_27_fc2_bias_to_fp16, dilations = hidden_states_57_dilations_0, groups = hidden_states_57_groups_0, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = hidden_states_57_strides_0, weight = layers_27_fc2_weight_to_fp16, x = input_279_cast_fp16)[name = tensor("hidden_states_57_cast_fp16")]; tensor inputs_169_cast_fp16 = add(x = inputs_167_cast_fp16, y = hidden_states_57_cast_fp16)[name = tensor("inputs_169_cast_fp16")]; tensor var_6356 = const()[name = tensor("op_6356"), val = tensor(3)]; tensor out_169_axes_0 = const()[name = tensor("out_169_axes_0"), val = tensor([1])]; tensor var_6381_to_fp16 = const()[name = tensor("op_6381_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_169_cast_fp16 = layer_norm(axes = out_169_axes_0, epsilon = var_6381_to_fp16, x = inputs_169_cast_fp16)[name = tensor("out_169_cast_fp16")]; tensor obj_393_gamma_0_to_fp16 = const()[name = tensor("obj_393_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1603194688)))]; tensor obj_393_beta_0_to_fp16 = const()[name = tensor("obj_393_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1603197312)))]; tensor obj_393_epsilon_0_to_fp16 = const()[name = tensor("obj_393_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_393_cast_fp16 = batch_norm(beta = obj_393_beta_0_to_fp16, epsilon = obj_393_epsilon_0_to_fp16, gamma = obj_393_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_169_cast_fp16)[name = tensor("obj_393_cast_fp16")]; tensor query_113_pad_type_0 = const()[name = tensor("query_113_pad_type_0"), val = tensor("valid")]; tensor query_113_strides_0 = const()[name = tensor("query_113_strides_0"), val = tensor([1, 1])]; tensor query_113_pad_0 = const()[name = tensor("query_113_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_113_dilations_0 = const()[name = tensor("query_113_dilations_0"), val = tensor([1, 1])]; tensor query_113_groups_0 = const()[name = tensor("query_113_groups_0"), val = tensor(1)]; tensor layers_28_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_28_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1603199936)))]; tensor layers_28_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_28_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1606476800)))]; tensor query_113_cast_fp16 = conv(bias = layers_28_self_attn_q_proj_bias_to_fp16, dilations = query_113_dilations_0, groups = query_113_groups_0, pad = query_113_pad_0, pad_type = query_113_pad_type_0, strides = query_113_strides_0, weight = layers_28_self_attn_q_proj_weight_to_fp16, x = obj_393_cast_fp16)[name = tensor("query_113_cast_fp16")]; tensor current_key_57_pad_type_0 = const()[name = tensor("current_key_57_pad_type_0"), val = tensor("valid")]; tensor current_key_57_strides_0 = const()[name = tensor("current_key_57_strides_0"), val = tensor([1, 1])]; tensor current_key_57_pad_0 = const()[name = tensor("current_key_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_57_dilations_0 = const()[name = tensor("current_key_57_dilations_0"), val = tensor([1, 1])]; tensor current_key_57_groups_0 = const()[name = tensor("current_key_57_groups_0"), val = tensor(1)]; tensor layers_28_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_28_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1606479424)))]; tensor current_key_57_cast_fp16 = conv(dilations = current_key_57_dilations_0, groups = current_key_57_groups_0, pad = current_key_57_pad_0, pad_type = current_key_57_pad_type_0, strides = current_key_57_strides_0, weight = layers_28_self_attn_k_proj_weight_to_fp16, x = obj_393_cast_fp16)[name = tensor("current_key_57_cast_fp16")]; tensor current_value_57_pad_type_0 = const()[name = tensor("current_value_57_pad_type_0"), val = tensor("valid")]; tensor current_value_57_strides_0 = const()[name = tensor("current_value_57_strides_0"), val = tensor([1, 1])]; tensor current_value_57_pad_0 = const()[name = tensor("current_value_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_57_dilations_0 = const()[name = tensor("current_value_57_dilations_0"), val = tensor([1, 1])]; tensor current_value_57_groups_0 = const()[name = tensor("current_value_57_groups_0"), val = tensor(1)]; tensor layers_28_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_28_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1609756288)))]; tensor layers_28_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_28_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1613033152)))]; tensor current_value_57_cast_fp16 = conv(bias = layers_28_self_attn_v_proj_bias_to_fp16, dilations = current_value_57_dilations_0, groups = current_value_57_groups_0, pad = current_value_57_pad_0, pad_type = current_value_57_pad_type_0, strides = current_value_57_strides_0, weight = layers_28_self_attn_v_proj_weight_to_fp16, x = obj_393_cast_fp16)[name = tensor("current_value_57_cast_fp16")]; tensor var_6420_cast_fp16 = mul(x = var_103_cast_fp16_28, y = var_239_cast_fp16)[name = tensor("op_6420_cast_fp16")]; tensor var_6421_cast_fp16 = mul(x = current_key_57_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_6421_cast_fp16")]; tensor key_113_cast_fp16 = add(x = var_6420_cast_fp16, y = var_6421_cast_fp16)[name = tensor("key_113_cast_fp16")]; tensor var_6424_cast_fp16 = mul(x = var_138_cast_fp16_28, y = var_239_cast_fp16)[name = tensor("op_6424_cast_fp16")]; tensor var_6425_cast_fp16 = mul(x = current_value_57_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_6425_cast_fp16")]; tensor value_113_cast_fp16 = add(x = var_6424_cast_fp16, y = var_6425_cast_fp16)[name = tensor("value_113_cast_fp16")]; tensor var_6429 = const()[name = tensor("op_6429"), val = tensor([1, 20, 64, 1])]; tensor mh_q_113_cast_fp16 = reshape(shape = var_6429, x = query_113_cast_fp16)[name = tensor("mh_q_113_cast_fp16")]; tensor var_6431_to_fp16 = const()[name = tensor("op_6431_to_fp16"), val = tensor(0x1p-3)]; tensor var_6432_cast_fp16 = mul(x = mh_q_113_cast_fp16, y = var_6431_to_fp16)[name = tensor("op_6432_cast_fp16")]; tensor var_6435 = const()[name = tensor("op_6435"), val = tensor([1, 20, 64, 448])]; tensor var_6436_cast_fp16 = reshape(shape = var_6435, x = key_113_cast_fp16)[name = tensor("op_6436_cast_fp16")]; tensor mh_w_169_transpose_x_0 = const()[name = tensor("mh_w_169_transpose_x_0"), val = tensor(true)]; tensor mh_w_169_transpose_y_0 = const()[name = tensor("mh_w_169_transpose_y_0"), val = tensor(false)]; tensor mh_w_169_cast_fp16 = matmul(transpose_x = mh_w_169_transpose_x_0, transpose_y = mh_w_169_transpose_y_0, x = var_6432_cast_fp16, y = var_6436_cast_fp16)[name = tensor("mh_w_169_cast_fp16")]; tensor mh_w_171_cast_fp16 = add(x = mh_w_169_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_171_cast_fp16")]; tensor var_6444_cast_fp16 = softmax(axis = var_6356, x = mh_w_171_cast_fp16)[name = tensor("op_6444_cast_fp16")]; tensor var_6445 = const()[name = tensor("op_6445"), val = tensor([1, 20, 64, 448])]; tensor var_6446_cast_fp16 = reshape(shape = var_6445, x = value_113_cast_fp16)[name = tensor("op_6446_cast_fp16")]; tensor attn_113_transpose_x_0 = const()[name = tensor("attn_113_transpose_x_0"), val = tensor(false)]; tensor attn_113_transpose_y_0 = const()[name = tensor("attn_113_transpose_y_0"), val = tensor(true)]; tensor attn_113_cast_fp16 = matmul(transpose_x = attn_113_transpose_x_0, transpose_y = attn_113_transpose_y_0, x = var_6446_cast_fp16, y = var_6444_cast_fp16)[name = tensor("attn_113_cast_fp16")]; tensor var_6449 = const()[name = tensor("op_6449"), val = tensor([1, 1280, 1, 1])]; tensor input_281_cast_fp16 = reshape(shape = var_6449, x = attn_113_cast_fp16)[name = tensor("input_281_cast_fp16")]; tensor obj_399_pad_type_0 = const()[name = tensor("obj_399_pad_type_0"), val = tensor("valid")]; tensor obj_399_strides_0 = const()[name = tensor("obj_399_strides_0"), val = tensor([1, 1])]; tensor obj_399_pad_0 = const()[name = tensor("obj_399_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_399_dilations_0 = const()[name = tensor("obj_399_dilations_0"), val = tensor([1, 1])]; tensor obj_399_groups_0 = const()[name = tensor("obj_399_groups_0"), val = tensor(1)]; tensor layers_28_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_28_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1613035776)))]; tensor layers_28_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_28_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1616312640)))]; tensor obj_399_cast_fp16 = conv(bias = layers_28_self_attn_o_proj_bias_to_fp16, dilations = obj_399_dilations_0, groups = obj_399_groups_0, pad = obj_399_pad_0, pad_type = obj_399_pad_type_0, strides = obj_399_strides_0, weight = layers_28_self_attn_o_proj_weight_to_fp16, x = input_281_cast_fp16)[name = tensor("obj_399_cast_fp16")]; tensor inputs_171_cast_fp16 = add(x = inputs_169_cast_fp16, y = obj_399_cast_fp16)[name = tensor("inputs_171_cast_fp16")]; tensor out_171_axes_0 = const()[name = tensor("out_171_axes_0"), val = tensor([1])]; tensor var_6471_to_fp16 = const()[name = tensor("op_6471_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_171_cast_fp16 = layer_norm(axes = out_171_axes_0, epsilon = var_6471_to_fp16, x = inputs_171_cast_fp16)[name = tensor("out_171_cast_fp16")]; tensor obj_401_gamma_0_to_fp16 = const()[name = tensor("obj_401_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1616315264)))]; tensor obj_401_beta_0_to_fp16 = const()[name = tensor("obj_401_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1616317888)))]; tensor obj_401_epsilon_0_to_fp16 = const()[name = tensor("obj_401_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_401_cast_fp16 = batch_norm(beta = obj_401_beta_0_to_fp16, epsilon = obj_401_epsilon_0_to_fp16, gamma = obj_401_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_171_cast_fp16)[name = tensor("obj_401_cast_fp16")]; tensor query_115_pad_type_0 = const()[name = tensor("query_115_pad_type_0"), val = tensor("valid")]; tensor query_115_strides_0 = const()[name = tensor("query_115_strides_0"), val = tensor([1, 1])]; tensor query_115_pad_0 = const()[name = tensor("query_115_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_115_dilations_0 = const()[name = tensor("query_115_dilations_0"), val = tensor([1, 1])]; tensor query_115_groups_0 = const()[name = tensor("query_115_groups_0"), val = tensor(1)]; tensor layers_28_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_28_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1616320512)))]; tensor layers_28_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_28_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1619597376)))]; tensor query_115_cast_fp16 = conv(bias = layers_28_encoder_attn_q_proj_bias_to_fp16, dilations = query_115_dilations_0, groups = query_115_groups_0, pad = query_115_pad_0, pad_type = query_115_pad_type_0, strides = query_115_strides_0, weight = layers_28_encoder_attn_q_proj_weight_to_fp16, x = obj_401_cast_fp16)[name = tensor("query_115_cast_fp16")]; tensor key_115_pad_type_0 = const()[name = tensor("key_115_pad_type_0"), val = tensor("valid")]; tensor key_115_strides_0 = const()[name = tensor("key_115_strides_0"), val = tensor([1, 1])]; tensor key_115_pad_0 = const()[name = tensor("key_115_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_115_dilations_0 = const()[name = tensor("key_115_dilations_0"), val = tensor([1, 1])]; tensor key_115_groups_0 = const()[name = tensor("key_115_groups_0"), val = tensor(1)]; tensor layers_28_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_28_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1619600000)))]; tensor key_115_cast_fp16 = conv(dilations = key_115_dilations_0, groups = key_115_groups_0, pad = key_115_pad_0, pad_type = key_115_pad_type_0, strides = key_115_strides_0, weight = layers_28_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_115_cast_fp16")]; tensor value_115_pad_type_0 = const()[name = tensor("value_115_pad_type_0"), val = tensor("valid")]; tensor value_115_strides_0 = const()[name = tensor("value_115_strides_0"), val = tensor([1, 1])]; tensor value_115_pad_0 = const()[name = tensor("value_115_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_115_dilations_0 = const()[name = tensor("value_115_dilations_0"), val = tensor([1, 1])]; tensor value_115_groups_0 = const()[name = tensor("value_115_groups_0"), val = tensor(1)]; tensor layers_28_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_28_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1622876864)))]; tensor layers_28_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_28_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1626153728)))]; tensor value_115_cast_fp16 = conv(bias = layers_28_encoder_attn_v_proj_bias_to_fp16, dilations = value_115_dilations_0, groups = value_115_groups_0, pad = value_115_pad_0, pad_type = value_115_pad_type_0, strides = value_115_strides_0, weight = layers_28_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_115_cast_fp16")]; tensor var_6507 = const()[name = tensor("op_6507"), val = tensor([1, 20, 64, 1])]; tensor mh_q_115_cast_fp16 = reshape(shape = var_6507, x = query_115_cast_fp16)[name = tensor("mh_q_115_cast_fp16")]; tensor var_6509_to_fp16 = const()[name = tensor("op_6509_to_fp16"), val = tensor(0x1p-3)]; tensor var_6510_cast_fp16 = mul(x = mh_q_115_cast_fp16, y = var_6509_to_fp16)[name = tensor("op_6510_cast_fp16")]; tensor var_6513 = const()[name = tensor("op_6513"), val = tensor([1, 20, 64, 1500])]; tensor var_6514_cast_fp16 = reshape(shape = var_6513, x = key_115_cast_fp16)[name = tensor("op_6514_cast_fp16")]; tensor mh_w_173_transpose_x_0 = const()[name = tensor("mh_w_173_transpose_x_0"), val = tensor(true)]; tensor mh_w_173_transpose_y_0 = const()[name = tensor("mh_w_173_transpose_y_0"), val = tensor(false)]; tensor mh_w_173_cast_fp16 = matmul(transpose_x = mh_w_173_transpose_x_0, transpose_y = mh_w_173_transpose_y_0, x = var_6510_cast_fp16, y = var_6514_cast_fp16)[name = tensor("mh_w_173_cast_fp16")]; tensor obj_405_cast_fp16 = softmax(axis = var_6356, x = mh_w_173_cast_fp16)[name = tensor("obj_405_cast_fp16")]; tensor var_6518 = const()[name = tensor("op_6518"), val = tensor([1, 20, 64, 1500])]; tensor var_6519_cast_fp16 = reshape(shape = var_6518, x = value_115_cast_fp16)[name = tensor("op_6519_cast_fp16")]; tensor attn_115_transpose_x_0 = const()[name = tensor("attn_115_transpose_x_0"), val = tensor(false)]; tensor attn_115_transpose_y_0 = const()[name = tensor("attn_115_transpose_y_0"), val = tensor(true)]; tensor attn_115_cast_fp16 = matmul(transpose_x = attn_115_transpose_x_0, transpose_y = attn_115_transpose_y_0, x = var_6519_cast_fp16, y = obj_405_cast_fp16)[name = tensor("attn_115_cast_fp16")]; tensor var_6522 = const()[name = tensor("op_6522"), val = tensor([1, 1280, 1, 1])]; tensor input_283_cast_fp16 = reshape(shape = var_6522, x = attn_115_cast_fp16)[name = tensor("input_283_cast_fp16")]; tensor obj_403_pad_type_0 = const()[name = tensor("obj_403_pad_type_0"), val = tensor("valid")]; tensor obj_403_strides_0 = const()[name = tensor("obj_403_strides_0"), val = tensor([1, 1])]; tensor obj_403_pad_0 = const()[name = tensor("obj_403_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_403_dilations_0 = const()[name = tensor("obj_403_dilations_0"), val = tensor([1, 1])]; tensor obj_403_groups_0 = const()[name = tensor("obj_403_groups_0"), val = tensor(1)]; tensor layers_28_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_28_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1626156352)))]; tensor layers_28_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_28_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1629433216)))]; tensor obj_403_cast_fp16 = conv(bias = layers_28_encoder_attn_o_proj_bias_to_fp16, dilations = obj_403_dilations_0, groups = obj_403_groups_0, pad = obj_403_pad_0, pad_type = obj_403_pad_type_0, strides = obj_403_strides_0, weight = layers_28_encoder_attn_o_proj_weight_to_fp16, x = input_283_cast_fp16)[name = tensor("obj_403_cast_fp16")]; tensor inputs_173_cast_fp16 = add(x = inputs_171_cast_fp16, y = obj_403_cast_fp16)[name = tensor("inputs_173_cast_fp16")]; tensor out_173_axes_0 = const()[name = tensor("out_173_axes_0"), val = tensor([1])]; tensor var_6540_to_fp16 = const()[name = tensor("op_6540_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_173_cast_fp16 = layer_norm(axes = out_173_axes_0, epsilon = var_6540_to_fp16, x = inputs_173_cast_fp16)[name = tensor("out_173_cast_fp16")]; tensor input_285_gamma_0_to_fp16 = const()[name = tensor("input_285_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1629435840)))]; tensor input_285_beta_0_to_fp16 = const()[name = tensor("input_285_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1629438464)))]; tensor input_285_epsilon_0_to_fp16 = const()[name = tensor("input_285_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_285_cast_fp16 = batch_norm(beta = input_285_beta_0_to_fp16, epsilon = input_285_epsilon_0_to_fp16, gamma = input_285_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_173_cast_fp16)[name = tensor("input_285_cast_fp16")]; tensor input_287_pad_type_0 = const()[name = tensor("input_287_pad_type_0"), val = tensor("valid")]; tensor input_287_strides_0 = const()[name = tensor("input_287_strides_0"), val = tensor([1, 1])]; tensor input_287_pad_0 = const()[name = tensor("input_287_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_287_dilations_0 = const()[name = tensor("input_287_dilations_0"), val = tensor([1, 1])]; tensor input_287_groups_0 = const()[name = tensor("input_287_groups_0"), val = tensor(1)]; tensor layers_28_fc1_weight_to_fp16 = const()[name = tensor("layers_28_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1629441088)))]; tensor layers_28_fc1_bias_to_fp16 = const()[name = tensor("layers_28_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1642548352)))]; tensor input_287_cast_fp16 = conv(bias = layers_28_fc1_bias_to_fp16, dilations = input_287_dilations_0, groups = input_287_groups_0, pad = input_287_pad_0, pad_type = input_287_pad_type_0, strides = input_287_strides_0, weight = layers_28_fc1_weight_to_fp16, x = input_285_cast_fp16)[name = tensor("input_287_cast_fp16")]; tensor input_289_mode_0 = const()[name = tensor("input_289_mode_0"), val = tensor("EXACT")]; tensor input_289_cast_fp16 = gelu(mode = input_289_mode_0, x = input_287_cast_fp16)[name = tensor("input_289_cast_fp16")]; tensor hidden_states_59_pad_type_0 = const()[name = tensor("hidden_states_59_pad_type_0"), val = tensor("valid")]; tensor hidden_states_59_strides_0 = const()[name = tensor("hidden_states_59_strides_0"), val = tensor([1, 1])]; tensor hidden_states_59_pad_0 = const()[name = tensor("hidden_states_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_59_dilations_0 = const()[name = tensor("hidden_states_59_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_59_groups_0 = const()[name = tensor("hidden_states_59_groups_0"), val = tensor(1)]; tensor layers_28_fc2_weight_to_fp16 = const()[name = tensor("layers_28_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1642558656)))]; tensor layers_28_fc2_bias_to_fp16 = const()[name = tensor("layers_28_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1655665920)))]; tensor hidden_states_59_cast_fp16 = conv(bias = layers_28_fc2_bias_to_fp16, dilations = hidden_states_59_dilations_0, groups = hidden_states_59_groups_0, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = hidden_states_59_strides_0, weight = layers_28_fc2_weight_to_fp16, x = input_289_cast_fp16)[name = tensor("hidden_states_59_cast_fp16")]; tensor inputs_175_cast_fp16 = add(x = inputs_173_cast_fp16, y = hidden_states_59_cast_fp16)[name = tensor("inputs_175_cast_fp16")]; tensor var_6575 = const()[name = tensor("op_6575"), val = tensor(3)]; tensor out_175_axes_0 = const()[name = tensor("out_175_axes_0"), val = tensor([1])]; tensor var_6600_to_fp16 = const()[name = tensor("op_6600_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_175_cast_fp16 = layer_norm(axes = out_175_axes_0, epsilon = var_6600_to_fp16, x = inputs_175_cast_fp16)[name = tensor("out_175_cast_fp16")]; tensor obj_407_gamma_0_to_fp16 = const()[name = tensor("obj_407_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1655668544)))]; tensor obj_407_beta_0_to_fp16 = const()[name = tensor("obj_407_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1655671168)))]; tensor obj_407_epsilon_0_to_fp16 = const()[name = tensor("obj_407_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_407_cast_fp16 = batch_norm(beta = obj_407_beta_0_to_fp16, epsilon = obj_407_epsilon_0_to_fp16, gamma = obj_407_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_175_cast_fp16)[name = tensor("obj_407_cast_fp16")]; tensor query_117_pad_type_0 = const()[name = tensor("query_117_pad_type_0"), val = tensor("valid")]; tensor query_117_strides_0 = const()[name = tensor("query_117_strides_0"), val = tensor([1, 1])]; tensor query_117_pad_0 = const()[name = tensor("query_117_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_117_dilations_0 = const()[name = tensor("query_117_dilations_0"), val = tensor([1, 1])]; tensor query_117_groups_0 = const()[name = tensor("query_117_groups_0"), val = tensor(1)]; tensor layers_29_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_29_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1655673792)))]; tensor layers_29_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_29_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1658950656)))]; tensor query_117_cast_fp16 = conv(bias = layers_29_self_attn_q_proj_bias_to_fp16, dilations = query_117_dilations_0, groups = query_117_groups_0, pad = query_117_pad_0, pad_type = query_117_pad_type_0, strides = query_117_strides_0, weight = layers_29_self_attn_q_proj_weight_to_fp16, x = obj_407_cast_fp16)[name = tensor("query_117_cast_fp16")]; tensor current_key_59_pad_type_0 = const()[name = tensor("current_key_59_pad_type_0"), val = tensor("valid")]; tensor current_key_59_strides_0 = const()[name = tensor("current_key_59_strides_0"), val = tensor([1, 1])]; tensor current_key_59_pad_0 = const()[name = tensor("current_key_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_59_dilations_0 = const()[name = tensor("current_key_59_dilations_0"), val = tensor([1, 1])]; tensor current_key_59_groups_0 = const()[name = tensor("current_key_59_groups_0"), val = tensor(1)]; tensor layers_29_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_29_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1658953280)))]; tensor current_key_59_cast_fp16 = conv(dilations = current_key_59_dilations_0, groups = current_key_59_groups_0, pad = current_key_59_pad_0, pad_type = current_key_59_pad_type_0, strides = current_key_59_strides_0, weight = layers_29_self_attn_k_proj_weight_to_fp16, x = obj_407_cast_fp16)[name = tensor("current_key_59_cast_fp16")]; tensor current_value_59_pad_type_0 = const()[name = tensor("current_value_59_pad_type_0"), val = tensor("valid")]; tensor current_value_59_strides_0 = const()[name = tensor("current_value_59_strides_0"), val = tensor([1, 1])]; tensor current_value_59_pad_0 = const()[name = tensor("current_value_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_59_dilations_0 = const()[name = tensor("current_value_59_dilations_0"), val = tensor([1, 1])]; tensor current_value_59_groups_0 = const()[name = tensor("current_value_59_groups_0"), val = tensor(1)]; tensor layers_29_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_29_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1662230144)))]; tensor layers_29_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_29_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1665507008)))]; tensor current_value_59_cast_fp16 = conv(bias = layers_29_self_attn_v_proj_bias_to_fp16, dilations = current_value_59_dilations_0, groups = current_value_59_groups_0, pad = current_value_59_pad_0, pad_type = current_value_59_pad_type_0, strides = current_value_59_strides_0, weight = layers_29_self_attn_v_proj_weight_to_fp16, x = obj_407_cast_fp16)[name = tensor("current_value_59_cast_fp16")]; tensor var_6639_cast_fp16 = mul(x = var_103_cast_fp16_29, y = var_239_cast_fp16)[name = tensor("op_6639_cast_fp16")]; tensor var_6640_cast_fp16 = mul(x = current_key_59_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_6640_cast_fp16")]; tensor key_117_cast_fp16 = add(x = var_6639_cast_fp16, y = var_6640_cast_fp16)[name = tensor("key_117_cast_fp16")]; tensor var_6643_cast_fp16 = mul(x = var_138_cast_fp16_29, y = var_239_cast_fp16)[name = tensor("op_6643_cast_fp16")]; tensor var_6644_cast_fp16 = mul(x = current_value_59_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_6644_cast_fp16")]; tensor value_117_cast_fp16 = add(x = var_6643_cast_fp16, y = var_6644_cast_fp16)[name = tensor("value_117_cast_fp16")]; tensor var_6648 = const()[name = tensor("op_6648"), val = tensor([1, 20, 64, 1])]; tensor mh_q_117_cast_fp16 = reshape(shape = var_6648, x = query_117_cast_fp16)[name = tensor("mh_q_117_cast_fp16")]; tensor var_6650_to_fp16 = const()[name = tensor("op_6650_to_fp16"), val = tensor(0x1p-3)]; tensor var_6651_cast_fp16 = mul(x = mh_q_117_cast_fp16, y = var_6650_to_fp16)[name = tensor("op_6651_cast_fp16")]; tensor var_6654 = const()[name = tensor("op_6654"), val = tensor([1, 20, 64, 448])]; tensor var_6655_cast_fp16 = reshape(shape = var_6654, x = key_117_cast_fp16)[name = tensor("op_6655_cast_fp16")]; tensor mh_w_175_transpose_x_0 = const()[name = tensor("mh_w_175_transpose_x_0"), val = tensor(true)]; tensor mh_w_175_transpose_y_0 = const()[name = tensor("mh_w_175_transpose_y_0"), val = tensor(false)]; tensor mh_w_175_cast_fp16 = matmul(transpose_x = mh_w_175_transpose_x_0, transpose_y = mh_w_175_transpose_y_0, x = var_6651_cast_fp16, y = var_6655_cast_fp16)[name = tensor("mh_w_175_cast_fp16")]; tensor mh_w_177_cast_fp16 = add(x = mh_w_175_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_177_cast_fp16")]; tensor var_6663_cast_fp16 = softmax(axis = var_6575, x = mh_w_177_cast_fp16)[name = tensor("op_6663_cast_fp16")]; tensor var_6664 = const()[name = tensor("op_6664"), val = tensor([1, 20, 64, 448])]; tensor var_6665_cast_fp16 = reshape(shape = var_6664, x = value_117_cast_fp16)[name = tensor("op_6665_cast_fp16")]; tensor attn_117_transpose_x_0 = const()[name = tensor("attn_117_transpose_x_0"), val = tensor(false)]; tensor attn_117_transpose_y_0 = const()[name = tensor("attn_117_transpose_y_0"), val = tensor(true)]; tensor attn_117_cast_fp16 = matmul(transpose_x = attn_117_transpose_x_0, transpose_y = attn_117_transpose_y_0, x = var_6665_cast_fp16, y = var_6663_cast_fp16)[name = tensor("attn_117_cast_fp16")]; tensor var_6668 = const()[name = tensor("op_6668"), val = tensor([1, 1280, 1, 1])]; tensor input_291_cast_fp16 = reshape(shape = var_6668, x = attn_117_cast_fp16)[name = tensor("input_291_cast_fp16")]; tensor obj_413_pad_type_0 = const()[name = tensor("obj_413_pad_type_0"), val = tensor("valid")]; tensor obj_413_strides_0 = const()[name = tensor("obj_413_strides_0"), val = tensor([1, 1])]; tensor obj_413_pad_0 = const()[name = tensor("obj_413_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_413_dilations_0 = const()[name = tensor("obj_413_dilations_0"), val = tensor([1, 1])]; tensor obj_413_groups_0 = const()[name = tensor("obj_413_groups_0"), val = tensor(1)]; tensor layers_29_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_29_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1665509632)))]; tensor layers_29_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_29_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1668786496)))]; tensor obj_413_cast_fp16 = conv(bias = layers_29_self_attn_o_proj_bias_to_fp16, dilations = obj_413_dilations_0, groups = obj_413_groups_0, pad = obj_413_pad_0, pad_type = obj_413_pad_type_0, strides = obj_413_strides_0, weight = layers_29_self_attn_o_proj_weight_to_fp16, x = input_291_cast_fp16)[name = tensor("obj_413_cast_fp16")]; tensor inputs_177_cast_fp16 = add(x = inputs_175_cast_fp16, y = obj_413_cast_fp16)[name = tensor("inputs_177_cast_fp16")]; tensor out_177_axes_0 = const()[name = tensor("out_177_axes_0"), val = tensor([1])]; tensor var_6690_to_fp16 = const()[name = tensor("op_6690_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_177_cast_fp16 = layer_norm(axes = out_177_axes_0, epsilon = var_6690_to_fp16, x = inputs_177_cast_fp16)[name = tensor("out_177_cast_fp16")]; tensor obj_415_gamma_0_to_fp16 = const()[name = tensor("obj_415_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1668789120)))]; tensor obj_415_beta_0_to_fp16 = const()[name = tensor("obj_415_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1668791744)))]; tensor obj_415_epsilon_0_to_fp16 = const()[name = tensor("obj_415_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_415_cast_fp16 = batch_norm(beta = obj_415_beta_0_to_fp16, epsilon = obj_415_epsilon_0_to_fp16, gamma = obj_415_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_177_cast_fp16)[name = tensor("obj_415_cast_fp16")]; tensor query_119_pad_type_0 = const()[name = tensor("query_119_pad_type_0"), val = tensor("valid")]; tensor query_119_strides_0 = const()[name = tensor("query_119_strides_0"), val = tensor([1, 1])]; tensor query_119_pad_0 = const()[name = tensor("query_119_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_119_dilations_0 = const()[name = tensor("query_119_dilations_0"), val = tensor([1, 1])]; tensor query_119_groups_0 = const()[name = tensor("query_119_groups_0"), val = tensor(1)]; tensor layers_29_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_29_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1668794368)))]; tensor layers_29_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_29_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1672071232)))]; tensor query_119_cast_fp16 = conv(bias = layers_29_encoder_attn_q_proj_bias_to_fp16, dilations = query_119_dilations_0, groups = query_119_groups_0, pad = query_119_pad_0, pad_type = query_119_pad_type_0, strides = query_119_strides_0, weight = layers_29_encoder_attn_q_proj_weight_to_fp16, x = obj_415_cast_fp16)[name = tensor("query_119_cast_fp16")]; tensor key_119_pad_type_0 = const()[name = tensor("key_119_pad_type_0"), val = tensor("valid")]; tensor key_119_strides_0 = const()[name = tensor("key_119_strides_0"), val = tensor([1, 1])]; tensor key_119_pad_0 = const()[name = tensor("key_119_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_119_dilations_0 = const()[name = tensor("key_119_dilations_0"), val = tensor([1, 1])]; tensor key_119_groups_0 = const()[name = tensor("key_119_groups_0"), val = tensor(1)]; tensor layers_29_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_29_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1672073856)))]; tensor key_119_cast_fp16 = conv(dilations = key_119_dilations_0, groups = key_119_groups_0, pad = key_119_pad_0, pad_type = key_119_pad_type_0, strides = key_119_strides_0, weight = layers_29_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_119_cast_fp16")]; tensor value_119_pad_type_0 = const()[name = tensor("value_119_pad_type_0"), val = tensor("valid")]; tensor value_119_strides_0 = const()[name = tensor("value_119_strides_0"), val = tensor([1, 1])]; tensor value_119_pad_0 = const()[name = tensor("value_119_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_119_dilations_0 = const()[name = tensor("value_119_dilations_0"), val = tensor([1, 1])]; tensor value_119_groups_0 = const()[name = tensor("value_119_groups_0"), val = tensor(1)]; tensor layers_29_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_29_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1675350720)))]; tensor layers_29_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_29_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1678627584)))]; tensor value_119_cast_fp16 = conv(bias = layers_29_encoder_attn_v_proj_bias_to_fp16, dilations = value_119_dilations_0, groups = value_119_groups_0, pad = value_119_pad_0, pad_type = value_119_pad_type_0, strides = value_119_strides_0, weight = layers_29_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_119_cast_fp16")]; tensor var_6726 = const()[name = tensor("op_6726"), val = tensor([1, 20, 64, 1])]; tensor mh_q_119_cast_fp16 = reshape(shape = var_6726, x = query_119_cast_fp16)[name = tensor("mh_q_119_cast_fp16")]; tensor var_6728_to_fp16 = const()[name = tensor("op_6728_to_fp16"), val = tensor(0x1p-3)]; tensor var_6729_cast_fp16 = mul(x = mh_q_119_cast_fp16, y = var_6728_to_fp16)[name = tensor("op_6729_cast_fp16")]; tensor var_6732 = const()[name = tensor("op_6732"), val = tensor([1, 20, 64, 1500])]; tensor var_6733_cast_fp16 = reshape(shape = var_6732, x = key_119_cast_fp16)[name = tensor("op_6733_cast_fp16")]; tensor mh_w_179_transpose_x_0 = const()[name = tensor("mh_w_179_transpose_x_0"), val = tensor(true)]; tensor mh_w_179_transpose_y_0 = const()[name = tensor("mh_w_179_transpose_y_0"), val = tensor(false)]; tensor mh_w_179_cast_fp16 = matmul(transpose_x = mh_w_179_transpose_x_0, transpose_y = mh_w_179_transpose_y_0, x = var_6729_cast_fp16, y = var_6733_cast_fp16)[name = tensor("mh_w_179_cast_fp16")]; tensor obj_419_cast_fp16 = softmax(axis = var_6575, x = mh_w_179_cast_fp16)[name = tensor("obj_419_cast_fp16")]; tensor var_6737 = const()[name = tensor("op_6737"), val = tensor([1, 20, 64, 1500])]; tensor var_6738_cast_fp16 = reshape(shape = var_6737, x = value_119_cast_fp16)[name = tensor("op_6738_cast_fp16")]; tensor attn_119_transpose_x_0 = const()[name = tensor("attn_119_transpose_x_0"), val = tensor(false)]; tensor attn_119_transpose_y_0 = const()[name = tensor("attn_119_transpose_y_0"), val = tensor(true)]; tensor attn_119_cast_fp16 = matmul(transpose_x = attn_119_transpose_x_0, transpose_y = attn_119_transpose_y_0, x = var_6738_cast_fp16, y = obj_419_cast_fp16)[name = tensor("attn_119_cast_fp16")]; tensor var_6741 = const()[name = tensor("op_6741"), val = tensor([1, 1280, 1, 1])]; tensor input_293_cast_fp16 = reshape(shape = var_6741, x = attn_119_cast_fp16)[name = tensor("input_293_cast_fp16")]; tensor obj_417_pad_type_0 = const()[name = tensor("obj_417_pad_type_0"), val = tensor("valid")]; tensor obj_417_strides_0 = const()[name = tensor("obj_417_strides_0"), val = tensor([1, 1])]; tensor obj_417_pad_0 = const()[name = tensor("obj_417_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_417_dilations_0 = const()[name = tensor("obj_417_dilations_0"), val = tensor([1, 1])]; tensor obj_417_groups_0 = const()[name = tensor("obj_417_groups_0"), val = tensor(1)]; tensor layers_29_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_29_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1678630208)))]; tensor layers_29_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_29_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1681907072)))]; tensor obj_417_cast_fp16 = conv(bias = layers_29_encoder_attn_o_proj_bias_to_fp16, dilations = obj_417_dilations_0, groups = obj_417_groups_0, pad = obj_417_pad_0, pad_type = obj_417_pad_type_0, strides = obj_417_strides_0, weight = layers_29_encoder_attn_o_proj_weight_to_fp16, x = input_293_cast_fp16)[name = tensor("obj_417_cast_fp16")]; tensor inputs_179_cast_fp16 = add(x = inputs_177_cast_fp16, y = obj_417_cast_fp16)[name = tensor("inputs_179_cast_fp16")]; tensor out_179_axes_0 = const()[name = tensor("out_179_axes_0"), val = tensor([1])]; tensor var_6759_to_fp16 = const()[name = tensor("op_6759_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_179_cast_fp16 = layer_norm(axes = out_179_axes_0, epsilon = var_6759_to_fp16, x = inputs_179_cast_fp16)[name = tensor("out_179_cast_fp16")]; tensor input_295_gamma_0_to_fp16 = const()[name = tensor("input_295_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1681909696)))]; tensor input_295_beta_0_to_fp16 = const()[name = tensor("input_295_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1681912320)))]; tensor input_295_epsilon_0_to_fp16 = const()[name = tensor("input_295_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_295_cast_fp16 = batch_norm(beta = input_295_beta_0_to_fp16, epsilon = input_295_epsilon_0_to_fp16, gamma = input_295_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_179_cast_fp16)[name = tensor("input_295_cast_fp16")]; tensor input_297_pad_type_0 = const()[name = tensor("input_297_pad_type_0"), val = tensor("valid")]; tensor input_297_strides_0 = const()[name = tensor("input_297_strides_0"), val = tensor([1, 1])]; tensor input_297_pad_0 = const()[name = tensor("input_297_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_297_dilations_0 = const()[name = tensor("input_297_dilations_0"), val = tensor([1, 1])]; tensor input_297_groups_0 = const()[name = tensor("input_297_groups_0"), val = tensor(1)]; tensor layers_29_fc1_weight_to_fp16 = const()[name = tensor("layers_29_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1681914944)))]; tensor layers_29_fc1_bias_to_fp16 = const()[name = tensor("layers_29_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1695022208)))]; tensor input_297_cast_fp16 = conv(bias = layers_29_fc1_bias_to_fp16, dilations = input_297_dilations_0, groups = input_297_groups_0, pad = input_297_pad_0, pad_type = input_297_pad_type_0, strides = input_297_strides_0, weight = layers_29_fc1_weight_to_fp16, x = input_295_cast_fp16)[name = tensor("input_297_cast_fp16")]; tensor input_299_mode_0 = const()[name = tensor("input_299_mode_0"), val = tensor("EXACT")]; tensor input_299_cast_fp16 = gelu(mode = input_299_mode_0, x = input_297_cast_fp16)[name = tensor("input_299_cast_fp16")]; tensor hidden_states_61_pad_type_0 = const()[name = tensor("hidden_states_61_pad_type_0"), val = tensor("valid")]; tensor hidden_states_61_strides_0 = const()[name = tensor("hidden_states_61_strides_0"), val = tensor([1, 1])]; tensor hidden_states_61_pad_0 = const()[name = tensor("hidden_states_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_61_dilations_0 = const()[name = tensor("hidden_states_61_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_61_groups_0 = const()[name = tensor("hidden_states_61_groups_0"), val = tensor(1)]; tensor layers_29_fc2_weight_to_fp16 = const()[name = tensor("layers_29_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1695032512)))]; tensor layers_29_fc2_bias_to_fp16 = const()[name = tensor("layers_29_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1708139776)))]; tensor hidden_states_61_cast_fp16 = conv(bias = layers_29_fc2_bias_to_fp16, dilations = hidden_states_61_dilations_0, groups = hidden_states_61_groups_0, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = hidden_states_61_strides_0, weight = layers_29_fc2_weight_to_fp16, x = input_299_cast_fp16)[name = tensor("hidden_states_61_cast_fp16")]; tensor inputs_181_cast_fp16 = add(x = inputs_179_cast_fp16, y = hidden_states_61_cast_fp16)[name = tensor("inputs_181_cast_fp16")]; tensor var_6794 = const()[name = tensor("op_6794"), val = tensor(3)]; tensor out_181_axes_0 = const()[name = tensor("out_181_axes_0"), val = tensor([1])]; tensor var_6819_to_fp16 = const()[name = tensor("op_6819_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_181_cast_fp16 = layer_norm(axes = out_181_axes_0, epsilon = var_6819_to_fp16, x = inputs_181_cast_fp16)[name = tensor("out_181_cast_fp16")]; tensor obj_421_gamma_0_to_fp16 = const()[name = tensor("obj_421_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1708142400)))]; tensor obj_421_beta_0_to_fp16 = const()[name = tensor("obj_421_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1708145024)))]; tensor obj_421_epsilon_0_to_fp16 = const()[name = tensor("obj_421_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_421_cast_fp16 = batch_norm(beta = obj_421_beta_0_to_fp16, epsilon = obj_421_epsilon_0_to_fp16, gamma = obj_421_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_181_cast_fp16)[name = tensor("obj_421_cast_fp16")]; tensor query_121_pad_type_0 = const()[name = tensor("query_121_pad_type_0"), val = tensor("valid")]; tensor query_121_strides_0 = const()[name = tensor("query_121_strides_0"), val = tensor([1, 1])]; tensor query_121_pad_0 = const()[name = tensor("query_121_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_121_dilations_0 = const()[name = tensor("query_121_dilations_0"), val = tensor([1, 1])]; tensor query_121_groups_0 = const()[name = tensor("query_121_groups_0"), val = tensor(1)]; tensor layers_30_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_30_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1708147648)))]; tensor layers_30_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_30_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1711424512)))]; tensor query_121_cast_fp16 = conv(bias = layers_30_self_attn_q_proj_bias_to_fp16, dilations = query_121_dilations_0, groups = query_121_groups_0, pad = query_121_pad_0, pad_type = query_121_pad_type_0, strides = query_121_strides_0, weight = layers_30_self_attn_q_proj_weight_to_fp16, x = obj_421_cast_fp16)[name = tensor("query_121_cast_fp16")]; tensor current_key_61_pad_type_0 = const()[name = tensor("current_key_61_pad_type_0"), val = tensor("valid")]; tensor current_key_61_strides_0 = const()[name = tensor("current_key_61_strides_0"), val = tensor([1, 1])]; tensor current_key_61_pad_0 = const()[name = tensor("current_key_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_61_dilations_0 = const()[name = tensor("current_key_61_dilations_0"), val = tensor([1, 1])]; tensor current_key_61_groups_0 = const()[name = tensor("current_key_61_groups_0"), val = tensor(1)]; tensor layers_30_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_30_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1711427136)))]; tensor current_key_61_cast_fp16 = conv(dilations = current_key_61_dilations_0, groups = current_key_61_groups_0, pad = current_key_61_pad_0, pad_type = current_key_61_pad_type_0, strides = current_key_61_strides_0, weight = layers_30_self_attn_k_proj_weight_to_fp16, x = obj_421_cast_fp16)[name = tensor("current_key_61_cast_fp16")]; tensor current_value_61_pad_type_0 = const()[name = tensor("current_value_61_pad_type_0"), val = tensor("valid")]; tensor current_value_61_strides_0 = const()[name = tensor("current_value_61_strides_0"), val = tensor([1, 1])]; tensor current_value_61_pad_0 = const()[name = tensor("current_value_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_61_dilations_0 = const()[name = tensor("current_value_61_dilations_0"), val = tensor([1, 1])]; tensor current_value_61_groups_0 = const()[name = tensor("current_value_61_groups_0"), val = tensor(1)]; tensor layers_30_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_30_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1714704000)))]; tensor layers_30_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_30_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1717980864)))]; tensor current_value_61_cast_fp16 = conv(bias = layers_30_self_attn_v_proj_bias_to_fp16, dilations = current_value_61_dilations_0, groups = current_value_61_groups_0, pad = current_value_61_pad_0, pad_type = current_value_61_pad_type_0, strides = current_value_61_strides_0, weight = layers_30_self_attn_v_proj_weight_to_fp16, x = obj_421_cast_fp16)[name = tensor("current_value_61_cast_fp16")]; tensor var_6858_cast_fp16 = mul(x = var_103_cast_fp16_30, y = var_239_cast_fp16)[name = tensor("op_6858_cast_fp16")]; tensor var_6859_cast_fp16 = mul(x = current_key_61_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_6859_cast_fp16")]; tensor key_121_cast_fp16 = add(x = var_6858_cast_fp16, y = var_6859_cast_fp16)[name = tensor("key_121_cast_fp16")]; tensor var_6862_cast_fp16 = mul(x = var_138_cast_fp16_30, y = var_239_cast_fp16)[name = tensor("op_6862_cast_fp16")]; tensor var_6863_cast_fp16 = mul(x = current_value_61_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_6863_cast_fp16")]; tensor value_121_cast_fp16 = add(x = var_6862_cast_fp16, y = var_6863_cast_fp16)[name = tensor("value_121_cast_fp16")]; tensor var_6867 = const()[name = tensor("op_6867"), val = tensor([1, 20, 64, 1])]; tensor mh_q_121_cast_fp16 = reshape(shape = var_6867, x = query_121_cast_fp16)[name = tensor("mh_q_121_cast_fp16")]; tensor var_6869_to_fp16 = const()[name = tensor("op_6869_to_fp16"), val = tensor(0x1p-3)]; tensor var_6870_cast_fp16 = mul(x = mh_q_121_cast_fp16, y = var_6869_to_fp16)[name = tensor("op_6870_cast_fp16")]; tensor var_6873 = const()[name = tensor("op_6873"), val = tensor([1, 20, 64, 448])]; tensor var_6874_cast_fp16 = reshape(shape = var_6873, x = key_121_cast_fp16)[name = tensor("op_6874_cast_fp16")]; tensor mh_w_181_transpose_x_0 = const()[name = tensor("mh_w_181_transpose_x_0"), val = tensor(true)]; tensor mh_w_181_transpose_y_0 = const()[name = tensor("mh_w_181_transpose_y_0"), val = tensor(false)]; tensor mh_w_181_cast_fp16 = matmul(transpose_x = mh_w_181_transpose_x_0, transpose_y = mh_w_181_transpose_y_0, x = var_6870_cast_fp16, y = var_6874_cast_fp16)[name = tensor("mh_w_181_cast_fp16")]; tensor mh_w_183_cast_fp16 = add(x = mh_w_181_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_183_cast_fp16")]; tensor var_6882_cast_fp16 = softmax(axis = var_6794, x = mh_w_183_cast_fp16)[name = tensor("op_6882_cast_fp16")]; tensor var_6883 = const()[name = tensor("op_6883"), val = tensor([1, 20, 64, 448])]; tensor var_6884_cast_fp16 = reshape(shape = var_6883, x = value_121_cast_fp16)[name = tensor("op_6884_cast_fp16")]; tensor attn_121_transpose_x_0 = const()[name = tensor("attn_121_transpose_x_0"), val = tensor(false)]; tensor attn_121_transpose_y_0 = const()[name = tensor("attn_121_transpose_y_0"), val = tensor(true)]; tensor attn_121_cast_fp16 = matmul(transpose_x = attn_121_transpose_x_0, transpose_y = attn_121_transpose_y_0, x = var_6884_cast_fp16, y = var_6882_cast_fp16)[name = tensor("attn_121_cast_fp16")]; tensor var_6887 = const()[name = tensor("op_6887"), val = tensor([1, 1280, 1, 1])]; tensor input_301_cast_fp16 = reshape(shape = var_6887, x = attn_121_cast_fp16)[name = tensor("input_301_cast_fp16")]; tensor obj_427_pad_type_0 = const()[name = tensor("obj_427_pad_type_0"), val = tensor("valid")]; tensor obj_427_strides_0 = const()[name = tensor("obj_427_strides_0"), val = tensor([1, 1])]; tensor obj_427_pad_0 = const()[name = tensor("obj_427_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_427_dilations_0 = const()[name = tensor("obj_427_dilations_0"), val = tensor([1, 1])]; tensor obj_427_groups_0 = const()[name = tensor("obj_427_groups_0"), val = tensor(1)]; tensor layers_30_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_30_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1717983488)))]; tensor layers_30_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_30_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1721260352)))]; tensor obj_427_cast_fp16 = conv(bias = layers_30_self_attn_o_proj_bias_to_fp16, dilations = obj_427_dilations_0, groups = obj_427_groups_0, pad = obj_427_pad_0, pad_type = obj_427_pad_type_0, strides = obj_427_strides_0, weight = layers_30_self_attn_o_proj_weight_to_fp16, x = input_301_cast_fp16)[name = tensor("obj_427_cast_fp16")]; tensor inputs_183_cast_fp16 = add(x = inputs_181_cast_fp16, y = obj_427_cast_fp16)[name = tensor("inputs_183_cast_fp16")]; tensor out_183_axes_0 = const()[name = tensor("out_183_axes_0"), val = tensor([1])]; tensor var_6909_to_fp16 = const()[name = tensor("op_6909_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_183_cast_fp16 = layer_norm(axes = out_183_axes_0, epsilon = var_6909_to_fp16, x = inputs_183_cast_fp16)[name = tensor("out_183_cast_fp16")]; tensor obj_429_gamma_0_to_fp16 = const()[name = tensor("obj_429_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1721262976)))]; tensor obj_429_beta_0_to_fp16 = const()[name = tensor("obj_429_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1721265600)))]; tensor obj_429_epsilon_0_to_fp16 = const()[name = tensor("obj_429_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_429_cast_fp16 = batch_norm(beta = obj_429_beta_0_to_fp16, epsilon = obj_429_epsilon_0_to_fp16, gamma = obj_429_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_183_cast_fp16)[name = tensor("obj_429_cast_fp16")]; tensor query_123_pad_type_0 = const()[name = tensor("query_123_pad_type_0"), val = tensor("valid")]; tensor query_123_strides_0 = const()[name = tensor("query_123_strides_0"), val = tensor([1, 1])]; tensor query_123_pad_0 = const()[name = tensor("query_123_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_123_dilations_0 = const()[name = tensor("query_123_dilations_0"), val = tensor([1, 1])]; tensor query_123_groups_0 = const()[name = tensor("query_123_groups_0"), val = tensor(1)]; tensor layers_30_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_30_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1721268224)))]; tensor layers_30_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_30_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1724545088)))]; tensor query_123_cast_fp16 = conv(bias = layers_30_encoder_attn_q_proj_bias_to_fp16, dilations = query_123_dilations_0, groups = query_123_groups_0, pad = query_123_pad_0, pad_type = query_123_pad_type_0, strides = query_123_strides_0, weight = layers_30_encoder_attn_q_proj_weight_to_fp16, x = obj_429_cast_fp16)[name = tensor("query_123_cast_fp16")]; tensor key_123_pad_type_0 = const()[name = tensor("key_123_pad_type_0"), val = tensor("valid")]; tensor key_123_strides_0 = const()[name = tensor("key_123_strides_0"), val = tensor([1, 1])]; tensor key_123_pad_0 = const()[name = tensor("key_123_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_123_dilations_0 = const()[name = tensor("key_123_dilations_0"), val = tensor([1, 1])]; tensor key_123_groups_0 = const()[name = tensor("key_123_groups_0"), val = tensor(1)]; tensor layers_30_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_30_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1724547712)))]; tensor key_123_cast_fp16 = conv(dilations = key_123_dilations_0, groups = key_123_groups_0, pad = key_123_pad_0, pad_type = key_123_pad_type_0, strides = key_123_strides_0, weight = layers_30_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_123_cast_fp16")]; tensor value_123_pad_type_0 = const()[name = tensor("value_123_pad_type_0"), val = tensor("valid")]; tensor value_123_strides_0 = const()[name = tensor("value_123_strides_0"), val = tensor([1, 1])]; tensor value_123_pad_0 = const()[name = tensor("value_123_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_123_dilations_0 = const()[name = tensor("value_123_dilations_0"), val = tensor([1, 1])]; tensor value_123_groups_0 = const()[name = tensor("value_123_groups_0"), val = tensor(1)]; tensor layers_30_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_30_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1727824576)))]; tensor layers_30_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_30_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1731101440)))]; tensor value_123_cast_fp16 = conv(bias = layers_30_encoder_attn_v_proj_bias_to_fp16, dilations = value_123_dilations_0, groups = value_123_groups_0, pad = value_123_pad_0, pad_type = value_123_pad_type_0, strides = value_123_strides_0, weight = layers_30_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_123_cast_fp16")]; tensor var_6945 = const()[name = tensor("op_6945"), val = tensor([1, 20, 64, 1])]; tensor mh_q_123_cast_fp16 = reshape(shape = var_6945, x = query_123_cast_fp16)[name = tensor("mh_q_123_cast_fp16")]; tensor var_6947_to_fp16 = const()[name = tensor("op_6947_to_fp16"), val = tensor(0x1p-3)]; tensor var_6948_cast_fp16 = mul(x = mh_q_123_cast_fp16, y = var_6947_to_fp16)[name = tensor("op_6948_cast_fp16")]; tensor var_6951 = const()[name = tensor("op_6951"), val = tensor([1, 20, 64, 1500])]; tensor var_6952_cast_fp16 = reshape(shape = var_6951, x = key_123_cast_fp16)[name = tensor("op_6952_cast_fp16")]; tensor mh_w_185_transpose_x_0 = const()[name = tensor("mh_w_185_transpose_x_0"), val = tensor(true)]; tensor mh_w_185_transpose_y_0 = const()[name = tensor("mh_w_185_transpose_y_0"), val = tensor(false)]; tensor mh_w_185_cast_fp16 = matmul(transpose_x = mh_w_185_transpose_x_0, transpose_y = mh_w_185_transpose_y_0, x = var_6948_cast_fp16, y = var_6952_cast_fp16)[name = tensor("mh_w_185_cast_fp16")]; tensor obj_433_cast_fp16 = softmax(axis = var_6794, x = mh_w_185_cast_fp16)[name = tensor("obj_433_cast_fp16")]; tensor var_6956 = const()[name = tensor("op_6956"), val = tensor([1, 20, 64, 1500])]; tensor var_6957_cast_fp16 = reshape(shape = var_6956, x = value_123_cast_fp16)[name = tensor("op_6957_cast_fp16")]; tensor attn_123_transpose_x_0 = const()[name = tensor("attn_123_transpose_x_0"), val = tensor(false)]; tensor attn_123_transpose_y_0 = const()[name = tensor("attn_123_transpose_y_0"), val = tensor(true)]; tensor attn_123_cast_fp16 = matmul(transpose_x = attn_123_transpose_x_0, transpose_y = attn_123_transpose_y_0, x = var_6957_cast_fp16, y = obj_433_cast_fp16)[name = tensor("attn_123_cast_fp16")]; tensor var_6960 = const()[name = tensor("op_6960"), val = tensor([1, 1280, 1, 1])]; tensor input_303_cast_fp16 = reshape(shape = var_6960, x = attn_123_cast_fp16)[name = tensor("input_303_cast_fp16")]; tensor obj_431_pad_type_0 = const()[name = tensor("obj_431_pad_type_0"), val = tensor("valid")]; tensor obj_431_strides_0 = const()[name = tensor("obj_431_strides_0"), val = tensor([1, 1])]; tensor obj_431_pad_0 = const()[name = tensor("obj_431_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_431_dilations_0 = const()[name = tensor("obj_431_dilations_0"), val = tensor([1, 1])]; tensor obj_431_groups_0 = const()[name = tensor("obj_431_groups_0"), val = tensor(1)]; tensor layers_30_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_30_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1731104064)))]; tensor layers_30_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_30_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1734380928)))]; tensor obj_431_cast_fp16 = conv(bias = layers_30_encoder_attn_o_proj_bias_to_fp16, dilations = obj_431_dilations_0, groups = obj_431_groups_0, pad = obj_431_pad_0, pad_type = obj_431_pad_type_0, strides = obj_431_strides_0, weight = layers_30_encoder_attn_o_proj_weight_to_fp16, x = input_303_cast_fp16)[name = tensor("obj_431_cast_fp16")]; tensor inputs_185_cast_fp16 = add(x = inputs_183_cast_fp16, y = obj_431_cast_fp16)[name = tensor("inputs_185_cast_fp16")]; tensor out_185_axes_0 = const()[name = tensor("out_185_axes_0"), val = tensor([1])]; tensor var_6978_to_fp16 = const()[name = tensor("op_6978_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_185_cast_fp16 = layer_norm(axes = out_185_axes_0, epsilon = var_6978_to_fp16, x = inputs_185_cast_fp16)[name = tensor("out_185_cast_fp16")]; tensor input_305_gamma_0_to_fp16 = const()[name = tensor("input_305_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1734383552)))]; tensor input_305_beta_0_to_fp16 = const()[name = tensor("input_305_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1734386176)))]; tensor input_305_epsilon_0_to_fp16 = const()[name = tensor("input_305_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_305_cast_fp16 = batch_norm(beta = input_305_beta_0_to_fp16, epsilon = input_305_epsilon_0_to_fp16, gamma = input_305_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_185_cast_fp16)[name = tensor("input_305_cast_fp16")]; tensor input_307_pad_type_0 = const()[name = tensor("input_307_pad_type_0"), val = tensor("valid")]; tensor input_307_strides_0 = const()[name = tensor("input_307_strides_0"), val = tensor([1, 1])]; tensor input_307_pad_0 = const()[name = tensor("input_307_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_307_dilations_0 = const()[name = tensor("input_307_dilations_0"), val = tensor([1, 1])]; tensor input_307_groups_0 = const()[name = tensor("input_307_groups_0"), val = tensor(1)]; tensor layers_30_fc1_weight_to_fp16 = const()[name = tensor("layers_30_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1734388800)))]; tensor layers_30_fc1_bias_to_fp16 = const()[name = tensor("layers_30_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1747496064)))]; tensor input_307_cast_fp16 = conv(bias = layers_30_fc1_bias_to_fp16, dilations = input_307_dilations_0, groups = input_307_groups_0, pad = input_307_pad_0, pad_type = input_307_pad_type_0, strides = input_307_strides_0, weight = layers_30_fc1_weight_to_fp16, x = input_305_cast_fp16)[name = tensor("input_307_cast_fp16")]; tensor input_309_mode_0 = const()[name = tensor("input_309_mode_0"), val = tensor("EXACT")]; tensor input_309_cast_fp16 = gelu(mode = input_309_mode_0, x = input_307_cast_fp16)[name = tensor("input_309_cast_fp16")]; tensor hidden_states_63_pad_type_0 = const()[name = tensor("hidden_states_63_pad_type_0"), val = tensor("valid")]; tensor hidden_states_63_strides_0 = const()[name = tensor("hidden_states_63_strides_0"), val = tensor([1, 1])]; tensor hidden_states_63_pad_0 = const()[name = tensor("hidden_states_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_63_dilations_0 = const()[name = tensor("hidden_states_63_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_63_groups_0 = const()[name = tensor("hidden_states_63_groups_0"), val = tensor(1)]; tensor layers_30_fc2_weight_to_fp16 = const()[name = tensor("layers_30_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1747506368)))]; tensor layers_30_fc2_bias_to_fp16 = const()[name = tensor("layers_30_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1760613632)))]; tensor hidden_states_63_cast_fp16 = conv(bias = layers_30_fc2_bias_to_fp16, dilations = hidden_states_63_dilations_0, groups = hidden_states_63_groups_0, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = hidden_states_63_strides_0, weight = layers_30_fc2_weight_to_fp16, x = input_309_cast_fp16)[name = tensor("hidden_states_63_cast_fp16")]; tensor inputs_187_cast_fp16 = add(x = inputs_185_cast_fp16, y = hidden_states_63_cast_fp16)[name = tensor("inputs_187_cast_fp16")]; tensor var_7013 = const()[name = tensor("op_7013"), val = tensor(3)]; tensor out_187_axes_0 = const()[name = tensor("out_187_axes_0"), val = tensor([1])]; tensor var_7038_to_fp16 = const()[name = tensor("op_7038_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_187_cast_fp16 = layer_norm(axes = out_187_axes_0, epsilon = var_7038_to_fp16, x = inputs_187_cast_fp16)[name = tensor("out_187_cast_fp16")]; tensor obj_435_gamma_0_to_fp16 = const()[name = tensor("obj_435_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1760616256)))]; tensor obj_435_beta_0_to_fp16 = const()[name = tensor("obj_435_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1760618880)))]; tensor obj_435_epsilon_0_to_fp16 = const()[name = tensor("obj_435_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_435_cast_fp16 = batch_norm(beta = obj_435_beta_0_to_fp16, epsilon = obj_435_epsilon_0_to_fp16, gamma = obj_435_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_187_cast_fp16)[name = tensor("obj_435_cast_fp16")]; tensor query_125_pad_type_0 = const()[name = tensor("query_125_pad_type_0"), val = tensor("valid")]; tensor query_125_strides_0 = const()[name = tensor("query_125_strides_0"), val = tensor([1, 1])]; tensor query_125_pad_0 = const()[name = tensor("query_125_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_125_dilations_0 = const()[name = tensor("query_125_dilations_0"), val = tensor([1, 1])]; tensor query_125_groups_0 = const()[name = tensor("query_125_groups_0"), val = tensor(1)]; tensor layers_31_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_31_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1760621504)))]; tensor layers_31_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_31_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1763898368)))]; tensor query_125_cast_fp16 = conv(bias = layers_31_self_attn_q_proj_bias_to_fp16, dilations = query_125_dilations_0, groups = query_125_groups_0, pad = query_125_pad_0, pad_type = query_125_pad_type_0, strides = query_125_strides_0, weight = layers_31_self_attn_q_proj_weight_to_fp16, x = obj_435_cast_fp16)[name = tensor("query_125_cast_fp16")]; tensor current_key_pad_type_0 = const()[name = tensor("current_key_pad_type_0"), val = tensor("valid")]; tensor current_key_strides_0 = const()[name = tensor("current_key_strides_0"), val = tensor([1, 1])]; tensor current_key_pad_0 = const()[name = tensor("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_dilations_0 = const()[name = tensor("current_key_dilations_0"), val = tensor([1, 1])]; tensor current_key_groups_0 = const()[name = tensor("current_key_groups_0"), val = tensor(1)]; tensor layers_31_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_31_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1763900992)))]; tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_31_self_attn_k_proj_weight_to_fp16, x = obj_435_cast_fp16)[name = tensor("current_key_cast_fp16")]; tensor current_value_pad_type_0 = const()[name = tensor("current_value_pad_type_0"), val = tensor("valid")]; tensor current_value_strides_0 = const()[name = tensor("current_value_strides_0"), val = tensor([1, 1])]; tensor current_value_pad_0 = const()[name = tensor("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_dilations_0 = const()[name = tensor("current_value_dilations_0"), val = tensor([1, 1])]; tensor current_value_groups_0 = const()[name = tensor("current_value_groups_0"), val = tensor(1)]; tensor layers_31_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_31_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1767177856)))]; tensor layers_31_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_31_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1770454720)))]; tensor current_value_cast_fp16 = conv(bias = layers_31_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_31_self_attn_v_proj_weight_to_fp16, x = obj_435_cast_fp16)[name = tensor("current_value_cast_fp16")]; tensor var_7077_cast_fp16 = mul(x = var_103_cast_fp16_31, y = var_239_cast_fp16)[name = tensor("op_7077_cast_fp16")]; tensor var_7078_cast_fp16 = mul(x = current_key_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_7078_cast_fp16")]; tensor key_125_cast_fp16 = add(x = var_7077_cast_fp16, y = var_7078_cast_fp16)[name = tensor("key_125_cast_fp16")]; tensor var_7081_cast_fp16 = mul(x = var_138_cast_fp16_31, y = var_239_cast_fp16)[name = tensor("op_7081_cast_fp16")]; tensor var_7082_cast_fp16 = mul(x = current_value_cast_fp16, y = var_237_cast_fp16)[name = tensor("op_7082_cast_fp16")]; tensor value_125_cast_fp16 = add(x = var_7081_cast_fp16, y = var_7082_cast_fp16)[name = tensor("value_125_cast_fp16")]; tensor var_7086 = const()[name = tensor("op_7086"), val = tensor([1, 20, 64, 1])]; tensor mh_q_125_cast_fp16 = reshape(shape = var_7086, x = query_125_cast_fp16)[name = tensor("mh_q_125_cast_fp16")]; tensor var_7088_to_fp16 = const()[name = tensor("op_7088_to_fp16"), val = tensor(0x1p-3)]; tensor var_7089_cast_fp16 = mul(x = mh_q_125_cast_fp16, y = var_7088_to_fp16)[name = tensor("op_7089_cast_fp16")]; tensor var_7092 = const()[name = tensor("op_7092"), val = tensor([1, 20, 64, 448])]; tensor var_7093_cast_fp16 = reshape(shape = var_7092, x = key_125_cast_fp16)[name = tensor("op_7093_cast_fp16")]; tensor mh_w_187_transpose_x_0 = const()[name = tensor("mh_w_187_transpose_x_0"), val = tensor(true)]; tensor mh_w_187_transpose_y_0 = const()[name = tensor("mh_w_187_transpose_y_0"), val = tensor(false)]; tensor mh_w_187_cast_fp16 = matmul(transpose_x = mh_w_187_transpose_x_0, transpose_y = mh_w_187_transpose_y_0, x = var_7089_cast_fp16, y = var_7093_cast_fp16)[name = tensor("mh_w_187_cast_fp16")]; tensor mh_w_189_cast_fp16 = add(x = mh_w_187_cast_fp16, y = var_261_cast_fp16)[name = tensor("mh_w_189_cast_fp16")]; tensor var_7101_cast_fp16 = softmax(axis = var_7013, x = mh_w_189_cast_fp16)[name = tensor("op_7101_cast_fp16")]; tensor var_7102 = const()[name = tensor("op_7102"), val = tensor([1, 20, 64, 448])]; tensor var_7103_cast_fp16 = reshape(shape = var_7102, x = value_125_cast_fp16)[name = tensor("op_7103_cast_fp16")]; tensor attn_125_transpose_x_0 = const()[name = tensor("attn_125_transpose_x_0"), val = tensor(false)]; tensor attn_125_transpose_y_0 = const()[name = tensor("attn_125_transpose_y_0"), val = tensor(true)]; tensor attn_125_cast_fp16 = matmul(transpose_x = attn_125_transpose_x_0, transpose_y = attn_125_transpose_y_0, x = var_7103_cast_fp16, y = var_7101_cast_fp16)[name = tensor("attn_125_cast_fp16")]; tensor var_7106 = const()[name = tensor("op_7106"), val = tensor([1, 1280, 1, 1])]; tensor input_311_cast_fp16 = reshape(shape = var_7106, x = attn_125_cast_fp16)[name = tensor("input_311_cast_fp16")]; tensor obj_441_pad_type_0 = const()[name = tensor("obj_441_pad_type_0"), val = tensor("valid")]; tensor obj_441_strides_0 = const()[name = tensor("obj_441_strides_0"), val = tensor([1, 1])]; tensor obj_441_pad_0 = const()[name = tensor("obj_441_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_441_dilations_0 = const()[name = tensor("obj_441_dilations_0"), val = tensor([1, 1])]; tensor obj_441_groups_0 = const()[name = tensor("obj_441_groups_0"), val = tensor(1)]; tensor layers_31_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_31_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1770457344)))]; tensor layers_31_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_31_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1773734208)))]; tensor obj_441_cast_fp16 = conv(bias = layers_31_self_attn_o_proj_bias_to_fp16, dilations = obj_441_dilations_0, groups = obj_441_groups_0, pad = obj_441_pad_0, pad_type = obj_441_pad_type_0, strides = obj_441_strides_0, weight = layers_31_self_attn_o_proj_weight_to_fp16, x = input_311_cast_fp16)[name = tensor("obj_441_cast_fp16")]; tensor inputs_189_cast_fp16 = add(x = inputs_187_cast_fp16, y = obj_441_cast_fp16)[name = tensor("inputs_189_cast_fp16")]; tensor out_189_axes_0 = const()[name = tensor("out_189_axes_0"), val = tensor([1])]; tensor var_7128_to_fp16 = const()[name = tensor("op_7128_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_189_cast_fp16 = layer_norm(axes = out_189_axes_0, epsilon = var_7128_to_fp16, x = inputs_189_cast_fp16)[name = tensor("out_189_cast_fp16")]; tensor obj_443_gamma_0_to_fp16 = const()[name = tensor("obj_443_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1773736832)))]; tensor obj_443_beta_0_to_fp16 = const()[name = tensor("obj_443_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1773739456)))]; tensor obj_443_epsilon_0_to_fp16 = const()[name = tensor("obj_443_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor obj_443_cast_fp16 = batch_norm(beta = obj_443_beta_0_to_fp16, epsilon = obj_443_epsilon_0_to_fp16, gamma = obj_443_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_189_cast_fp16)[name = tensor("obj_443_cast_fp16")]; tensor query_pad_type_0 = const()[name = tensor("query_pad_type_0"), val = tensor("valid")]; tensor query_strides_0 = const()[name = tensor("query_strides_0"), val = tensor([1, 1])]; tensor query_pad_0 = const()[name = tensor("query_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_dilations_0 = const()[name = tensor("query_dilations_0"), val = tensor([1, 1])]; tensor query_groups_0 = const()[name = tensor("query_groups_0"), val = tensor(1)]; tensor layers_31_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_31_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1773742080)))]; tensor layers_31_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_31_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1777018944)))]; tensor query_cast_fp16 = conv(bias = layers_31_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_31_encoder_attn_q_proj_weight_to_fp16, x = obj_443_cast_fp16)[name = tensor("query_cast_fp16")]; tensor key_pad_type_0 = const()[name = tensor("key_pad_type_0"), val = tensor("valid")]; tensor key_strides_0 = const()[name = tensor("key_strides_0"), val = tensor([1, 1])]; tensor key_pad_0 = const()[name = tensor("key_pad_0"), val = tensor([0, 0, 0, 0])]; tensor key_dilations_0 = const()[name = tensor("key_dilations_0"), val = tensor([1, 1])]; tensor key_groups_0 = const()[name = tensor("key_groups_0"), val = tensor(1)]; tensor layers_31_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_31_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1777021568)))]; tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_31_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("key_cast_fp16")]; tensor value_pad_type_0 = const()[name = tensor("value_pad_type_0"), val = tensor("valid")]; tensor value_strides_0 = const()[name = tensor("value_strides_0"), val = tensor([1, 1])]; tensor value_pad_0 = const()[name = tensor("value_pad_0"), val = tensor([0, 0, 0, 0])]; tensor value_dilations_0 = const()[name = tensor("value_dilations_0"), val = tensor([1, 1])]; tensor value_groups_0 = const()[name = tensor("value_groups_0"), val = tensor(1)]; tensor layers_31_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_31_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1780298432)))]; tensor layers_31_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_31_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1783575296)))]; tensor value_cast_fp16 = conv(bias = layers_31_encoder_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_31_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor("value_cast_fp16")]; tensor var_7164 = const()[name = tensor("op_7164"), val = tensor([1, 20, 64, 1])]; tensor mh_q_cast_fp16 = reshape(shape = var_7164, x = query_cast_fp16)[name = tensor("mh_q_cast_fp16")]; tensor var_7166_to_fp16 = const()[name = tensor("op_7166_to_fp16"), val = tensor(0x1p-3)]; tensor var_7167_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_7166_to_fp16)[name = tensor("op_7167_cast_fp16")]; tensor var_7170 = const()[name = tensor("op_7170"), val = tensor([1, 20, 64, 1500])]; tensor var_7171_cast_fp16 = reshape(shape = var_7170, x = key_cast_fp16)[name = tensor("op_7171_cast_fp16")]; tensor mh_w_transpose_x_0 = const()[name = tensor("mh_w_transpose_x_0"), val = tensor(true)]; tensor mh_w_transpose_y_0 = const()[name = tensor("mh_w_transpose_y_0"), val = tensor(false)]; tensor mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_7167_cast_fp16, y = var_7171_cast_fp16)[name = tensor("mh_w_cast_fp16")]; tensor obj_447_cast_fp16 = softmax(axis = var_7013, x = mh_w_cast_fp16)[name = tensor("obj_447_cast_fp16")]; tensor var_7175 = const()[name = tensor("op_7175"), val = tensor([1, 20, 64, 1500])]; tensor var_7176_cast_fp16 = reshape(shape = var_7175, x = value_cast_fp16)[name = tensor("op_7176_cast_fp16")]; tensor attn_transpose_x_0 = const()[name = tensor("attn_transpose_x_0"), val = tensor(false)]; tensor attn_transpose_y_0 = const()[name = tensor("attn_transpose_y_0"), val = tensor(true)]; tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_7176_cast_fp16, y = obj_447_cast_fp16)[name = tensor("attn_cast_fp16")]; tensor var_7179 = const()[name = tensor("op_7179"), val = tensor([1, 1280, 1, 1])]; tensor input_313_cast_fp16 = reshape(shape = var_7179, x = attn_cast_fp16)[name = tensor("input_313_cast_fp16")]; tensor obj_445_pad_type_0 = const()[name = tensor("obj_445_pad_type_0"), val = tensor("valid")]; tensor obj_445_strides_0 = const()[name = tensor("obj_445_strides_0"), val = tensor([1, 1])]; tensor obj_445_pad_0 = const()[name = tensor("obj_445_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_445_dilations_0 = const()[name = tensor("obj_445_dilations_0"), val = tensor([1, 1])]; tensor obj_445_groups_0 = const()[name = tensor("obj_445_groups_0"), val = tensor(1)]; tensor layers_31_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_31_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1783577920)))]; tensor layers_31_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_31_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1786854784)))]; tensor obj_445_cast_fp16 = conv(bias = layers_31_encoder_attn_o_proj_bias_to_fp16, dilations = obj_445_dilations_0, groups = obj_445_groups_0, pad = obj_445_pad_0, pad_type = obj_445_pad_type_0, strides = obj_445_strides_0, weight = layers_31_encoder_attn_o_proj_weight_to_fp16, x = input_313_cast_fp16)[name = tensor("obj_445_cast_fp16")]; tensor inputs_191_cast_fp16 = add(x = inputs_189_cast_fp16, y = obj_445_cast_fp16)[name = tensor("inputs_191_cast_fp16")]; tensor out_191_axes_0 = const()[name = tensor("out_191_axes_0"), val = tensor([1])]; tensor var_7197_to_fp16 = const()[name = tensor("op_7197_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_191_cast_fp16 = layer_norm(axes = out_191_axes_0, epsilon = var_7197_to_fp16, x = inputs_191_cast_fp16)[name = tensor("out_191_cast_fp16")]; tensor input_315_gamma_0_to_fp16 = const()[name = tensor("input_315_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1786857408)))]; tensor input_315_beta_0_to_fp16 = const()[name = tensor("input_315_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1786860032)))]; tensor input_315_epsilon_0_to_fp16 = const()[name = tensor("input_315_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_315_cast_fp16 = batch_norm(beta = input_315_beta_0_to_fp16, epsilon = input_315_epsilon_0_to_fp16, gamma = input_315_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_191_cast_fp16)[name = tensor("input_315_cast_fp16")]; tensor input_317_pad_type_0 = const()[name = tensor("input_317_pad_type_0"), val = tensor("valid")]; tensor input_317_strides_0 = const()[name = tensor("input_317_strides_0"), val = tensor([1, 1])]; tensor input_317_pad_0 = const()[name = tensor("input_317_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_317_dilations_0 = const()[name = tensor("input_317_dilations_0"), val = tensor([1, 1])]; tensor input_317_groups_0 = const()[name = tensor("input_317_groups_0"), val = tensor(1)]; tensor layers_31_fc1_weight_to_fp16 = const()[name = tensor("layers_31_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1786862656)))]; tensor layers_31_fc1_bias_to_fp16 = const()[name = tensor("layers_31_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1799969920)))]; tensor input_317_cast_fp16 = conv(bias = layers_31_fc1_bias_to_fp16, dilations = input_317_dilations_0, groups = input_317_groups_0, pad = input_317_pad_0, pad_type = input_317_pad_type_0, strides = input_317_strides_0, weight = layers_31_fc1_weight_to_fp16, x = input_315_cast_fp16)[name = tensor("input_317_cast_fp16")]; tensor input_mode_0 = const()[name = tensor("input_mode_0"), val = tensor("EXACT")]; tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_317_cast_fp16)[name = tensor("input_cast_fp16")]; tensor hidden_states_65_pad_type_0 = const()[name = tensor("hidden_states_65_pad_type_0"), val = tensor("valid")]; tensor hidden_states_65_strides_0 = const()[name = tensor("hidden_states_65_strides_0"), val = tensor([1, 1])]; tensor hidden_states_65_pad_0 = const()[name = tensor("hidden_states_65_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_65_dilations_0 = const()[name = tensor("hidden_states_65_dilations_0"), val = tensor([1, 1])]; tensor hidden_states_65_groups_0 = const()[name = tensor("hidden_states_65_groups_0"), val = tensor(1)]; tensor layers_31_fc2_weight_to_fp16 = const()[name = tensor("layers_31_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1799980224)))]; tensor layers_31_fc2_bias_to_fp16 = const()[name = tensor("layers_31_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1813087488)))]; tensor hidden_states_65_cast_fp16 = conv(bias = layers_31_fc2_bias_to_fp16, dilations = hidden_states_65_dilations_0, groups = hidden_states_65_groups_0, pad = hidden_states_65_pad_0, pad_type = hidden_states_65_pad_type_0, strides = hidden_states_65_strides_0, weight = layers_31_fc2_weight_to_fp16, x = input_cast_fp16)[name = tensor("hidden_states_65_cast_fp16")]; tensor inputs_cast_fp16 = add(x = inputs_191_cast_fp16, y = hidden_states_65_cast_fp16)[name = tensor("inputs_cast_fp16")]; tensor out_axes_0 = const()[name = tensor("out_axes_0"), val = tensor([1])]; tensor var_7239_to_fp16 = const()[name = tensor("op_7239_to_fp16"), val = tensor(0x1.5p-17)]; tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_7239_to_fp16, x = inputs_cast_fp16)[name = tensor("out_cast_fp16")]; tensor hidden_states_gamma_0_to_fp16 = const()[name = tensor("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1813090112)))]; tensor hidden_states_beta_0_to_fp16 = const()[name = tensor("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1813092736)))]; tensor hidden_states_epsilon_0_to_fp16 = const()[name = tensor("hidden_states_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = tensor("hidden_states_cast_fp16")]; tensor var_7250_axes_0 = const()[name = tensor("op_7250_axes_0"), val = tensor([2])]; tensor var_7250_cast_fp16 = squeeze(axes = var_7250_axes_0, x = hidden_states_cast_fp16)[name = tensor("op_7250_cast_fp16")]; tensor var_7253_perm_0 = const()[name = tensor("op_7253_perm_0"), val = tensor([0, 2, 1])]; tensor linear_0_bias_0_to_fp16 = const()[name = tensor("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1813095360)))]; tensor var_7253_cast_fp16 = transpose(perm = var_7253_perm_0, x = var_7250_cast_fp16)[name = tensor("transpose_0")]; tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_7253_cast_fp16)[name = tensor("linear_0_cast_fp16")]; tensor var_7257 = const()[name = tensor("op_7257"), val = tensor(1)]; tensor obj_451_interleave_0 = const()[name = tensor("obj_451_interleave_0"), val = tensor(false)]; tensor key_cache_updates = concat(axis = var_7257, interleave = obj_451_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_7_cast_fp16, current_key_9_cast_fp16, current_key_11_cast_fp16, current_key_13_cast_fp16, current_key_15_cast_fp16, current_key_17_cast_fp16, current_key_19_cast_fp16, current_key_21_cast_fp16, current_key_23_cast_fp16, current_key_25_cast_fp16, current_key_27_cast_fp16, current_key_29_cast_fp16, current_key_31_cast_fp16, current_key_33_cast_fp16, current_key_35_cast_fp16, current_key_37_cast_fp16, current_key_39_cast_fp16, current_key_41_cast_fp16, current_key_43_cast_fp16, current_key_45_cast_fp16, current_key_47_cast_fp16, current_key_49_cast_fp16, current_key_51_cast_fp16, current_key_53_cast_fp16, current_key_55_cast_fp16, current_key_57_cast_fp16, current_key_59_cast_fp16, current_key_61_cast_fp16, current_key_cast_fp16))[name = tensor("obj_451_cast_fp16")]; tensor var_7260 = const()[name = tensor("op_7260"), val = tensor(1)]; tensor obj_453_interleave_0 = const()[name = tensor("obj_453_interleave_0"), val = tensor(false)]; tensor value_cache_updates = concat(axis = var_7260, interleave = obj_453_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_7_cast_fp16, current_value_9_cast_fp16, current_value_11_cast_fp16, current_value_13_cast_fp16, current_value_15_cast_fp16, current_value_17_cast_fp16, current_value_19_cast_fp16, current_value_21_cast_fp16, current_value_23_cast_fp16, current_value_25_cast_fp16, current_value_27_cast_fp16, current_value_29_cast_fp16, current_value_31_cast_fp16, current_value_33_cast_fp16, current_value_35_cast_fp16, current_value_37_cast_fp16, current_value_39_cast_fp16, current_value_41_cast_fp16, current_value_43_cast_fp16, current_value_45_cast_fp16, current_value_47_cast_fp16, current_value_49_cast_fp16, current_value_51_cast_fp16, current_value_53_cast_fp16, current_value_55_cast_fp16, current_value_57_cast_fp16, current_value_59_cast_fp16, current_value_61_cast_fp16, current_value_cast_fp16))[name = tensor("obj_453_cast_fp16")]; tensor var_7271_begin_0 = const()[name = tensor("op_7271_begin_0"), val = tensor([0, 12, 0, 0])]; tensor var_7271_end_0 = const()[name = tensor("op_7271_end_0"), val = tensor([1, 13, 1, 1500])]; tensor var_7271_end_mask_0 = const()[name = tensor("op_7271_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7271_cast_fp16 = slice_by_index(begin = var_7271_begin_0, end = var_7271_end_0, end_mask = var_7271_end_mask_0, x = obj_153_cast_fp16)[name = tensor("op_7271_cast_fp16")]; tensor var_7274_begin_0 = const()[name = tensor("op_7274_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7274_end_0 = const()[name = tensor("op_7274_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7274_end_mask_0 = const()[name = tensor("op_7274_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7274_squeeze_mask_0 = const()[name = tensor("op_7274_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7274_cast_fp16 = slice_by_index(begin = var_7274_begin_0, end = var_7274_end_0, end_mask = var_7274_end_mask_0, squeeze_mask = var_7274_squeeze_mask_0, x = var_7271_cast_fp16)[name = tensor("op_7274_cast_fp16")]; tensor var_7289_begin_0 = const()[name = tensor("op_7289_begin_0"), val = tensor([0, 17, 0, 0])]; tensor var_7289_end_0 = const()[name = tensor("op_7289_end_0"), val = tensor([1, 18, 1, 1500])]; tensor var_7289_end_mask_0 = const()[name = tensor("op_7289_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7289_cast_fp16 = slice_by_index(begin = var_7289_begin_0, end = var_7289_end_0, end_mask = var_7289_end_mask_0, x = obj_195_cast_fp16)[name = tensor("op_7289_cast_fp16")]; tensor var_7292_begin_0 = const()[name = tensor("op_7292_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7292_end_0 = const()[name = tensor("op_7292_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7292_end_mask_0 = const()[name = tensor("op_7292_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7292_squeeze_mask_0 = const()[name = tensor("op_7292_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7292_cast_fp16 = slice_by_index(begin = var_7292_begin_0, end = var_7292_end_0, end_mask = var_7292_end_mask_0, squeeze_mask = var_7292_squeeze_mask_0, x = var_7289_cast_fp16)[name = tensor("op_7292_cast_fp16")]; tensor var_7307_begin_0 = const()[name = tensor("op_7307_begin_0"), val = tensor([0, 11, 0, 0])]; tensor var_7307_end_0 = const()[name = tensor("op_7307_end_0"), val = tensor([1, 12, 1, 1500])]; tensor var_7307_end_mask_0 = const()[name = tensor("op_7307_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7307_cast_fp16 = slice_by_index(begin = var_7307_begin_0, end = var_7307_end_0, end_mask = var_7307_end_mask_0, x = obj_237_cast_fp16)[name = tensor("op_7307_cast_fp16")]; tensor var_7310_begin_0 = const()[name = tensor("op_7310_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7310_end_0 = const()[name = tensor("op_7310_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7310_end_mask_0 = const()[name = tensor("op_7310_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7310_squeeze_mask_0 = const()[name = tensor("op_7310_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7310_cast_fp16 = slice_by_index(begin = var_7310_begin_0, end = var_7310_end_0, end_mask = var_7310_end_mask_0, squeeze_mask = var_7310_squeeze_mask_0, x = var_7307_cast_fp16)[name = tensor("op_7310_cast_fp16")]; tensor var_7325_begin_0 = const()[name = tensor("op_7325_begin_0"), val = tensor([0, 12, 0, 0])]; tensor var_7325_end_0 = const()[name = tensor("op_7325_end_0"), val = tensor([1, 13, 1, 1500])]; tensor var_7325_end_mask_0 = const()[name = tensor("op_7325_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7325_cast_fp16 = slice_by_index(begin = var_7325_begin_0, end = var_7325_end_0, end_mask = var_7325_end_mask_0, x = obj_237_cast_fp16)[name = tensor("op_7325_cast_fp16")]; tensor var_7328_begin_0 = const()[name = tensor("op_7328_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7328_end_0 = const()[name = tensor("op_7328_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7328_end_mask_0 = const()[name = tensor("op_7328_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7328_squeeze_mask_0 = const()[name = tensor("op_7328_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7328_cast_fp16 = slice_by_index(begin = var_7328_begin_0, end = var_7328_end_0, end_mask = var_7328_end_mask_0, squeeze_mask = var_7328_squeeze_mask_0, x = var_7325_cast_fp16)[name = tensor("op_7328_cast_fp16")]; tensor var_7343_begin_0 = const()[name = tensor("op_7343_begin_0"), val = tensor([0, 13, 0, 0])]; tensor var_7343_end_0 = const()[name = tensor("op_7343_end_0"), val = tensor([1, 14, 1, 1500])]; tensor var_7343_end_mask_0 = const()[name = tensor("op_7343_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7343_cast_fp16 = slice_by_index(begin = var_7343_begin_0, end = var_7343_end_0, end_mask = var_7343_end_mask_0, x = obj_237_cast_fp16)[name = tensor("op_7343_cast_fp16")]; tensor var_7346_begin_0 = const()[name = tensor("op_7346_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7346_end_0 = const()[name = tensor("op_7346_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7346_end_mask_0 = const()[name = tensor("op_7346_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7346_squeeze_mask_0 = const()[name = tensor("op_7346_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7346_cast_fp16 = slice_by_index(begin = var_7346_begin_0, end = var_7346_end_0, end_mask = var_7346_end_mask_0, squeeze_mask = var_7346_squeeze_mask_0, x = var_7343_cast_fp16)[name = tensor("op_7346_cast_fp16")]; tensor var_7361_begin_0 = const()[name = tensor("op_7361_begin_0"), val = tensor([0, 15, 0, 0])]; tensor var_7361_end_0 = const()[name = tensor("op_7361_end_0"), val = tensor([1, 16, 1, 1500])]; tensor var_7361_end_mask_0 = const()[name = tensor("op_7361_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7361_cast_fp16 = slice_by_index(begin = var_7361_begin_0, end = var_7361_end_0, end_mask = var_7361_end_mask_0, x = obj_251_cast_fp16)[name = tensor("op_7361_cast_fp16")]; tensor var_7364_begin_0 = const()[name = tensor("op_7364_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7364_end_0 = const()[name = tensor("op_7364_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7364_end_mask_0 = const()[name = tensor("op_7364_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7364_squeeze_mask_0 = const()[name = tensor("op_7364_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7364_cast_fp16 = slice_by_index(begin = var_7364_begin_0, end = var_7364_end_0, end_mask = var_7364_end_mask_0, squeeze_mask = var_7364_squeeze_mask_0, x = var_7361_cast_fp16)[name = tensor("op_7364_cast_fp16")]; tensor var_7379_begin_0 = const()[name = tensor("op_7379_begin_0"), val = tensor([0, 16, 0, 0])]; tensor var_7379_end_0 = const()[name = tensor("op_7379_end_0"), val = tensor([1, 17, 1, 1500])]; tensor var_7379_end_mask_0 = const()[name = tensor("op_7379_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7379_cast_fp16 = slice_by_index(begin = var_7379_begin_0, end = var_7379_end_0, end_mask = var_7379_end_mask_0, x = obj_251_cast_fp16)[name = tensor("op_7379_cast_fp16")]; tensor var_7382_begin_0 = const()[name = tensor("op_7382_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7382_end_0 = const()[name = tensor("op_7382_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7382_end_mask_0 = const()[name = tensor("op_7382_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7382_squeeze_mask_0 = const()[name = tensor("op_7382_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7382_cast_fp16 = slice_by_index(begin = var_7382_begin_0, end = var_7382_end_0, end_mask = var_7382_end_mask_0, squeeze_mask = var_7382_squeeze_mask_0, x = var_7379_cast_fp16)[name = tensor("op_7382_cast_fp16")]; tensor var_7397_begin_0 = const()[name = tensor("op_7397_begin_0"), val = tensor([0, 4, 0, 0])]; tensor var_7397_end_0 = const()[name = tensor("op_7397_end_0"), val = tensor([1, 5, 1, 1500])]; tensor var_7397_end_mask_0 = const()[name = tensor("op_7397_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7397_cast_fp16 = slice_by_index(begin = var_7397_begin_0, end = var_7397_end_0, end_mask = var_7397_end_mask_0, x = obj_265_cast_fp16)[name = tensor("op_7397_cast_fp16")]; tensor var_7400_begin_0 = const()[name = tensor("op_7400_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7400_end_0 = const()[name = tensor("op_7400_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7400_end_mask_0 = const()[name = tensor("op_7400_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7400_squeeze_mask_0 = const()[name = tensor("op_7400_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7400_cast_fp16 = slice_by_index(begin = var_7400_begin_0, end = var_7400_end_0, end_mask = var_7400_end_mask_0, squeeze_mask = var_7400_squeeze_mask_0, x = var_7397_cast_fp16)[name = tensor("op_7400_cast_fp16")]; tensor var_7415_begin_0 = const()[name = tensor("op_7415_begin_0"), val = tensor([0, 11, 0, 0])]; tensor var_7415_end_0 = const()[name = tensor("op_7415_end_0"), val = tensor([1, 12, 1, 1500])]; tensor var_7415_end_mask_0 = const()[name = tensor("op_7415_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7415_cast_fp16 = slice_by_index(begin = var_7415_begin_0, end = var_7415_end_0, end_mask = var_7415_end_mask_0, x = obj_265_cast_fp16)[name = tensor("op_7415_cast_fp16")]; tensor var_7418_begin_0 = const()[name = tensor("op_7418_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7418_end_0 = const()[name = tensor("op_7418_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7418_end_mask_0 = const()[name = tensor("op_7418_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7418_squeeze_mask_0 = const()[name = tensor("op_7418_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7418_cast_fp16 = slice_by_index(begin = var_7418_begin_0, end = var_7418_end_0, end_mask = var_7418_end_mask_0, squeeze_mask = var_7418_squeeze_mask_0, x = var_7415_cast_fp16)[name = tensor("op_7418_cast_fp16")]; tensor var_7433_begin_0 = const()[name = tensor("op_7433_begin_0"), val = tensor([0, 19, 0, 0])]; tensor var_7433_end_0 = const()[name = tensor("op_7433_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7433_end_mask_0 = const()[name = tensor("op_7433_end_mask_0"), val = tensor([true, true, true, true])]; tensor var_7433_cast_fp16 = slice_by_index(begin = var_7433_begin_0, end = var_7433_end_0, end_mask = var_7433_end_mask_0, x = obj_265_cast_fp16)[name = tensor("op_7433_cast_fp16")]; tensor var_7436_begin_0 = const()[name = tensor("op_7436_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7436_end_0 = const()[name = tensor("op_7436_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7436_end_mask_0 = const()[name = tensor("op_7436_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7436_squeeze_mask_0 = const()[name = tensor("op_7436_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7436_cast_fp16 = slice_by_index(begin = var_7436_begin_0, end = var_7436_end_0, end_mask = var_7436_end_mask_0, squeeze_mask = var_7436_squeeze_mask_0, x = var_7433_cast_fp16)[name = tensor("op_7436_cast_fp16")]; tensor var_7451_begin_0 = const()[name = tensor("op_7451_begin_0"), val = tensor([0, 11, 0, 0])]; tensor var_7451_end_0 = const()[name = tensor("op_7451_end_0"), val = tensor([1, 12, 1, 1500])]; tensor var_7451_end_mask_0 = const()[name = tensor("op_7451_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7451_cast_fp16 = slice_by_index(begin = var_7451_begin_0, end = var_7451_end_0, end_mask = var_7451_end_mask_0, x = obj_279_cast_fp16)[name = tensor("op_7451_cast_fp16")]; tensor var_7454_begin_0 = const()[name = tensor("op_7454_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7454_end_0 = const()[name = tensor("op_7454_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7454_end_mask_0 = const()[name = tensor("op_7454_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7454_squeeze_mask_0 = const()[name = tensor("op_7454_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7454_cast_fp16 = slice_by_index(begin = var_7454_begin_0, end = var_7454_end_0, end_mask = var_7454_end_mask_0, squeeze_mask = var_7454_squeeze_mask_0, x = var_7451_cast_fp16)[name = tensor("op_7454_cast_fp16")]; tensor var_7469_begin_0 = const()[name = tensor("op_7469_begin_0"), val = tensor([0, 2, 0, 0])]; tensor var_7469_end_0 = const()[name = tensor("op_7469_end_0"), val = tensor([1, 3, 1, 1500])]; tensor var_7469_end_mask_0 = const()[name = tensor("op_7469_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7469_cast_fp16 = slice_by_index(begin = var_7469_begin_0, end = var_7469_end_0, end_mask = var_7469_end_mask_0, x = obj_307_cast_fp16)[name = tensor("op_7469_cast_fp16")]; tensor var_7472_begin_0 = const()[name = tensor("op_7472_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7472_end_0 = const()[name = tensor("op_7472_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7472_end_mask_0 = const()[name = tensor("op_7472_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7472_squeeze_mask_0 = const()[name = tensor("op_7472_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7472_cast_fp16 = slice_by_index(begin = var_7472_begin_0, end = var_7472_end_0, end_mask = var_7472_end_mask_0, squeeze_mask = var_7472_squeeze_mask_0, x = var_7469_cast_fp16)[name = tensor("op_7472_cast_fp16")]; tensor var_7487_begin_0 = const()[name = tensor("op_7487_begin_0"), val = tensor([0, 3, 0, 0])]; tensor var_7487_end_0 = const()[name = tensor("op_7487_end_0"), val = tensor([1, 4, 1, 1500])]; tensor var_7487_end_mask_0 = const()[name = tensor("op_7487_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7487_cast_fp16 = slice_by_index(begin = var_7487_begin_0, end = var_7487_end_0, end_mask = var_7487_end_mask_0, x = obj_307_cast_fp16)[name = tensor("op_7487_cast_fp16")]; tensor var_7490_begin_0 = const()[name = tensor("op_7490_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7490_end_0 = const()[name = tensor("op_7490_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7490_end_mask_0 = const()[name = tensor("op_7490_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7490_squeeze_mask_0 = const()[name = tensor("op_7490_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7490_cast_fp16 = slice_by_index(begin = var_7490_begin_0, end = var_7490_end_0, end_mask = var_7490_end_mask_0, squeeze_mask = var_7490_squeeze_mask_0, x = var_7487_cast_fp16)[name = tensor("op_7490_cast_fp16")]; tensor var_7505_begin_0 = const()[name = tensor("op_7505_begin_0"), val = tensor([0, 3, 0, 0])]; tensor var_7505_end_0 = const()[name = tensor("op_7505_end_0"), val = tensor([1, 4, 1, 1500])]; tensor var_7505_end_mask_0 = const()[name = tensor("op_7505_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7505_cast_fp16 = slice_by_index(begin = var_7505_begin_0, end = var_7505_end_0, end_mask = var_7505_end_mask_0, x = obj_321_cast_fp16)[name = tensor("op_7505_cast_fp16")]; tensor var_7508_begin_0 = const()[name = tensor("op_7508_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7508_end_0 = const()[name = tensor("op_7508_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7508_end_mask_0 = const()[name = tensor("op_7508_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7508_squeeze_mask_0 = const()[name = tensor("op_7508_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7508_cast_fp16 = slice_by_index(begin = var_7508_begin_0, end = var_7508_end_0, end_mask = var_7508_end_mask_0, squeeze_mask = var_7508_squeeze_mask_0, x = var_7505_cast_fp16)[name = tensor("op_7508_cast_fp16")]; tensor var_7523_begin_0 = const()[name = tensor("op_7523_begin_0"), val = tensor([0, 9, 0, 0])]; tensor var_7523_end_0 = const()[name = tensor("op_7523_end_0"), val = tensor([1, 10, 1, 1500])]; tensor var_7523_end_mask_0 = const()[name = tensor("op_7523_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7523_cast_fp16 = slice_by_index(begin = var_7523_begin_0, end = var_7523_end_0, end_mask = var_7523_end_mask_0, x = obj_321_cast_fp16)[name = tensor("op_7523_cast_fp16")]; tensor var_7526_begin_0 = const()[name = tensor("op_7526_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7526_end_0 = const()[name = tensor("op_7526_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7526_end_mask_0 = const()[name = tensor("op_7526_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7526_squeeze_mask_0 = const()[name = tensor("op_7526_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7526_cast_fp16 = slice_by_index(begin = var_7526_begin_0, end = var_7526_end_0, end_mask = var_7526_end_mask_0, squeeze_mask = var_7526_squeeze_mask_0, x = var_7523_cast_fp16)[name = tensor("op_7526_cast_fp16")]; tensor var_7541_begin_0 = const()[name = tensor("op_7541_begin_0"), val = tensor([0, 12, 0, 0])]; tensor var_7541_end_0 = const()[name = tensor("op_7541_end_0"), val = tensor([1, 13, 1, 1500])]; tensor var_7541_end_mask_0 = const()[name = tensor("op_7541_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7541_cast_fp16 = slice_by_index(begin = var_7541_begin_0, end = var_7541_end_0, end_mask = var_7541_end_mask_0, x = obj_321_cast_fp16)[name = tensor("op_7541_cast_fp16")]; tensor var_7544_begin_0 = const()[name = tensor("op_7544_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7544_end_0 = const()[name = tensor("op_7544_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7544_end_mask_0 = const()[name = tensor("op_7544_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7544_squeeze_mask_0 = const()[name = tensor("op_7544_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7544_cast_fp16 = slice_by_index(begin = var_7544_begin_0, end = var_7544_end_0, end_mask = var_7544_end_mask_0, squeeze_mask = var_7544_squeeze_mask_0, x = var_7541_cast_fp16)[name = tensor("op_7544_cast_fp16")]; tensor var_7559_begin_0 = const()[name = tensor("op_7559_begin_0"), val = tensor([0, 5, 0, 0])]; tensor var_7559_end_0 = const()[name = tensor("op_7559_end_0"), val = tensor([1, 6, 1, 1500])]; tensor var_7559_end_mask_0 = const()[name = tensor("op_7559_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7559_cast_fp16 = slice_by_index(begin = var_7559_begin_0, end = var_7559_end_0, end_mask = var_7559_end_mask_0, x = obj_335_cast_fp16)[name = tensor("op_7559_cast_fp16")]; tensor var_7562_begin_0 = const()[name = tensor("op_7562_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7562_end_0 = const()[name = tensor("op_7562_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7562_end_mask_0 = const()[name = tensor("op_7562_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7562_squeeze_mask_0 = const()[name = tensor("op_7562_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7562_cast_fp16 = slice_by_index(begin = var_7562_begin_0, end = var_7562_end_0, end_mask = var_7562_end_mask_0, squeeze_mask = var_7562_squeeze_mask_0, x = var_7559_cast_fp16)[name = tensor("op_7562_cast_fp16")]; tensor var_7577_begin_0 = const()[name = tensor("op_7577_begin_0"), val = tensor([0, 7, 0, 0])]; tensor var_7577_end_0 = const()[name = tensor("op_7577_end_0"), val = tensor([1, 8, 1, 1500])]; tensor var_7577_end_mask_0 = const()[name = tensor("op_7577_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7577_cast_fp16 = slice_by_index(begin = var_7577_begin_0, end = var_7577_end_0, end_mask = var_7577_end_mask_0, x = obj_335_cast_fp16)[name = tensor("op_7577_cast_fp16")]; tensor var_7580_begin_0 = const()[name = tensor("op_7580_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7580_end_0 = const()[name = tensor("op_7580_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7580_end_mask_0 = const()[name = tensor("op_7580_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7580_squeeze_mask_0 = const()[name = tensor("op_7580_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7580_cast_fp16 = slice_by_index(begin = var_7580_begin_0, end = var_7580_end_0, end_mask = var_7580_end_mask_0, squeeze_mask = var_7580_squeeze_mask_0, x = var_7577_cast_fp16)[name = tensor("op_7580_cast_fp16")]; tensor var_7595_begin_0 = const()[name = tensor("op_7595_begin_0"), val = tensor([0, 13, 0, 0])]; tensor var_7595_end_0 = const()[name = tensor("op_7595_end_0"), val = tensor([1, 14, 1, 1500])]; tensor var_7595_end_mask_0 = const()[name = tensor("op_7595_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7595_cast_fp16 = slice_by_index(begin = var_7595_begin_0, end = var_7595_end_0, end_mask = var_7595_end_mask_0, x = obj_335_cast_fp16)[name = tensor("op_7595_cast_fp16")]; tensor var_7598_begin_0 = const()[name = tensor("op_7598_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7598_end_0 = const()[name = tensor("op_7598_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7598_end_mask_0 = const()[name = tensor("op_7598_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7598_squeeze_mask_0 = const()[name = tensor("op_7598_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7598_cast_fp16 = slice_by_index(begin = var_7598_begin_0, end = var_7598_end_0, end_mask = var_7598_end_mask_0, squeeze_mask = var_7598_squeeze_mask_0, x = var_7595_cast_fp16)[name = tensor("op_7598_cast_fp16")]; tensor var_7613_begin_0 = const()[name = tensor("op_7613_begin_0"), val = tensor([0, 5, 0, 0])]; tensor var_7613_end_0 = const()[name = tensor("op_7613_end_0"), val = tensor([1, 6, 1, 1500])]; tensor var_7613_end_mask_0 = const()[name = tensor("op_7613_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7613_cast_fp16 = slice_by_index(begin = var_7613_begin_0, end = var_7613_end_0, end_mask = var_7613_end_mask_0, x = obj_363_cast_fp16)[name = tensor("op_7613_cast_fp16")]; tensor var_7616_begin_0 = const()[name = tensor("op_7616_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7616_end_0 = const()[name = tensor("op_7616_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7616_end_mask_0 = const()[name = tensor("op_7616_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7616_squeeze_mask_0 = const()[name = tensor("op_7616_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7616_cast_fp16 = slice_by_index(begin = var_7616_begin_0, end = var_7616_end_0, end_mask = var_7616_end_mask_0, squeeze_mask = var_7616_squeeze_mask_0, x = var_7613_cast_fp16)[name = tensor("op_7616_cast_fp16")]; tensor var_7631_begin_0 = const()[name = tensor("op_7631_begin_0"), val = tensor([0, 1, 0, 0])]; tensor var_7631_end_0 = const()[name = tensor("op_7631_end_0"), val = tensor([1, 2, 1, 1500])]; tensor var_7631_end_mask_0 = const()[name = tensor("op_7631_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7631_cast_fp16 = slice_by_index(begin = var_7631_begin_0, end = var_7631_end_0, end_mask = var_7631_end_mask_0, x = obj_377_cast_fp16)[name = tensor("op_7631_cast_fp16")]; tensor var_7634_begin_0 = const()[name = tensor("op_7634_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7634_end_0 = const()[name = tensor("op_7634_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7634_end_mask_0 = const()[name = tensor("op_7634_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7634_squeeze_mask_0 = const()[name = tensor("op_7634_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7634_cast_fp16 = slice_by_index(begin = var_7634_begin_0, end = var_7634_end_0, end_mask = var_7634_end_mask_0, squeeze_mask = var_7634_squeeze_mask_0, x = var_7631_cast_fp16)[name = tensor("op_7634_cast_fp16")]; tensor var_7649_begin_0 = const()[name = tensor("op_7649_begin_0"), val = tensor([0, 12, 0, 0])]; tensor var_7649_end_0 = const()[name = tensor("op_7649_end_0"), val = tensor([1, 13, 1, 1500])]; tensor var_7649_end_mask_0 = const()[name = tensor("op_7649_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7649_cast_fp16 = slice_by_index(begin = var_7649_begin_0, end = var_7649_end_0, end_mask = var_7649_end_mask_0, x = obj_377_cast_fp16)[name = tensor("op_7649_cast_fp16")]; tensor var_7652_begin_0 = const()[name = tensor("op_7652_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7652_end_0 = const()[name = tensor("op_7652_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7652_end_mask_0 = const()[name = tensor("op_7652_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7652_squeeze_mask_0 = const()[name = tensor("op_7652_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7652_cast_fp16 = slice_by_index(begin = var_7652_begin_0, end = var_7652_end_0, end_mask = var_7652_end_mask_0, squeeze_mask = var_7652_squeeze_mask_0, x = var_7649_cast_fp16)[name = tensor("op_7652_cast_fp16")]; tensor var_7667_begin_0 = const()[name = tensor("op_7667_begin_0"), val = tensor([0, 15, 0, 0])]; tensor var_7667_end_0 = const()[name = tensor("op_7667_end_0"), val = tensor([1, 16, 1, 1500])]; tensor var_7667_end_mask_0 = const()[name = tensor("op_7667_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7667_cast_fp16 = slice_by_index(begin = var_7667_begin_0, end = var_7667_end_0, end_mask = var_7667_end_mask_0, x = obj_391_cast_fp16)[name = tensor("op_7667_cast_fp16")]; tensor var_7670_begin_0 = const()[name = tensor("op_7670_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7670_end_0 = const()[name = tensor("op_7670_end_0"), val = tensor([1, 1, 1, 1500])]; tensor var_7670_end_mask_0 = const()[name = tensor("op_7670_end_mask_0"), val = tensor([true, true, false, true])]; tensor var_7670_squeeze_mask_0 = const()[name = tensor("op_7670_squeeze_mask_0"), val = tensor([false, false, true, false])]; tensor var_7670_cast_fp16 = slice_by_index(begin = var_7670_begin_0, end = var_7670_end_0, end_mask = var_7670_end_mask_0, squeeze_mask = var_7670_squeeze_mask_0, x = var_7667_cast_fp16)[name = tensor("op_7670_cast_fp16")]; tensor var_7677 = const()[name = tensor("op_7677"), val = tensor(1)]; tensor var_7678_interleave_0 = const()[name = tensor("op_7678_interleave_0"), val = tensor(false)]; tensor var_7678_cast_fp16 = concat(axis = var_7677, interleave = var_7678_interleave_0, values = (var_7274_cast_fp16, var_7292_cast_fp16, var_7310_cast_fp16, var_7328_cast_fp16, var_7346_cast_fp16, var_7364_cast_fp16, var_7382_cast_fp16, var_7400_cast_fp16, var_7418_cast_fp16, var_7436_cast_fp16, var_7454_cast_fp16, var_7472_cast_fp16, var_7490_cast_fp16, var_7508_cast_fp16, var_7526_cast_fp16, var_7544_cast_fp16, var_7562_cast_fp16, var_7580_cast_fp16, var_7598_cast_fp16, var_7616_cast_fp16, var_7634_cast_fp16, var_7652_cast_fp16, var_7670_cast_fp16))[name = tensor("op_7678_cast_fp16")]; tensor obj_axes_0 = const()[name = tensor("obj_axes_0"), val = tensor([1])]; tensor obj_keep_dims_0 = const()[name = tensor("obj_keep_dims_0"), val = tensor(false)]; tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = obj_keep_dims_0, x = var_7678_cast_fp16)[name = tensor("obj_cast_fp16")]; } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); }