program(1.3) [buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] { func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { int32 var_26_axis_0 = const()[name = string("op_26_axis_0"), val = int32(0)]; int32 var_26_batch_dims_0 = const()[name = string("op_26_batch_dims_0"), val = int32(0)]; bool var_26_validate_indices_0 = const()[name = string("op_26_validate_indices_0"), val = bool(false)]; tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; tensor var_26_cast_fp16 = gather(axis = var_26_axis_0, batch_dims = var_26_batch_dims_0, indices = input_ids, validate_indices = var_26_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_26_cast_fp16")]; int32 var_30_axis_0 = const()[name = string("op_30_axis_0"), val = int32(0)]; int32 var_30_batch_dims_0 = const()[name = string("op_30_batch_dims_0"), val = int32(0)]; bool var_30_validate_indices_0 = const()[name = string("op_30_validate_indices_0"), val = bool(false)]; tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132777088)))]; string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_71")]; tensor var_30_cast_fp16_cast_uint16 = gather(axis = var_30_axis_0, batch_dims = var_30_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_30_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_30_cast_fp16_cast_uint16")]; tensor hidden_states_1_cast_fp16 = add(x = var_26_cast_fp16, y = var_30_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; tensor var_44_axes_0 = const()[name = string("op_44_axes_0"), val = tensor([2])]; tensor var_44_cast_fp16 = expand_dims(axes = var_44_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_44_cast_fp16")]; tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_44_cast_fp16)[name = string("inputs_1_cast_fp16")]; tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; tensor tile_0 = const()[name = string("tile_0"), val = tensor([1, 1, 1, 1])]; int32 var_49_axis_0 = const()[name = string("op_49_axis_0"), val = int32(0)]; tensor var_49_cast_fp16_0, tensor var_49_cast_fp16_1, tensor var_49_cast_fp16_2, tensor var_49_cast_fp16_3 = split(axis = var_49_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_49_cast_fp16")]; tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; tensor tile_1 = const()[name = string("tile_1"), val = tensor([1, 1, 1, 1])]; int32 var_56_axis_0 = const()[name = string("op_56_axis_0"), val = int32(0)]; tensor var_56_cast_fp16_0, tensor var_56_cast_fp16_1, tensor var_56_cast_fp16_2, tensor var_56_cast_fp16_3 = split(axis = var_56_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_56_cast_fp16")]; tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 1280, 1, 1536])]; tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 1280, 1, 1536])]; tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; int32 var_76 = const()[name = string("op_76"), val = int32(3)]; tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; fp16 var_101_to_fp16 = const()[name = string("op_101_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_101_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133924032)))]; tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133926656)))]; tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133929280)))]; tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133931904)))]; fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133934528)))]; tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137211392)))]; 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_5_cast_fp16)[name = string("query_1_cast_fp16")]; string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137214016)))]; 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_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140490880)))]; tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143767744)))]; 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_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; tensor var_136_axes_0 = const()[name = string("op_136_axes_0"), val = tensor([1])]; tensor var_136_cast_fp16 = expand_dims(axes = var_136_axes_0, x = kv_cache_update_mask)[name = string("op_136_cast_fp16")]; tensor var_137_axes_0 = const()[name = string("op_137_axes_0"), val = tensor([2])]; tensor var_137_cast_fp16 = expand_dims(axes = var_137_axes_0, x = var_136_cast_fp16)[name = string("op_137_cast_fp16")]; tensor var_139_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_139_cast_fp16")]; tensor key_1_cast_fp16 = add(x = var_49_cast_fp16_0, y = var_139_cast_fp16)[name = string("key_1_cast_fp16")]; tensor var_141_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_141_cast_fp16")]; tensor value_1_cast_fp16 = add(x = var_56_cast_fp16_0, y = var_141_cast_fp16)[name = string("value_1_cast_fp16")]; tensor var_144 = const()[name = string("op_144"), val = tensor([1, 20, 64, -1])]; tensor mh_q_1_cast_fp16 = reshape(shape = var_144, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; fp16 var_146_to_fp16 = const()[name = string("op_146_to_fp16"), val = fp16(0x1p-3)]; tensor var_147_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_146_to_fp16)[name = string("op_147_cast_fp16")]; tensor var_148 = const()[name = string("op_148"), val = tensor([1, 20, 64, -1])]; tensor var_149_cast_fp16 = reshape(shape = var_148, x = key_1_cast_fp16)[name = string("op_149_cast_fp16")]; bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(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_147_cast_fp16, y = var_149_cast_fp16)[name = string("mh_w_1_cast_fp16")]; tensor var_153_axes_0 = const()[name = string("op_153_axes_0"), val = tensor([1])]; tensor var_153_cast_fp16 = expand_dims(axes = var_153_axes_0, x = decoder_key_padding_mask)[name = string("op_153_cast_fp16")]; tensor var_154_axes_0 = const()[name = string("op_154_axes_0"), val = tensor([2])]; tensor var_154_cast_fp16 = expand_dims(axes = var_154_axes_0, x = var_153_cast_fp16)[name = string("op_154_cast_fp16")]; tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_3_cast_fp16")]; tensor var_157_cast_fp16 = softmax(axis = var_76, x = mh_w_3_cast_fp16)[name = string("op_157_cast_fp16")]; tensor var_158 = const()[name = string("op_158"), val = tensor([1, 20, 64, -1])]; tensor var_159_cast_fp16 = reshape(shape = var_158, x = value_1_cast_fp16)[name = string("op_159_cast_fp16")]; bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_159_cast_fp16, y = var_157_cast_fp16)[name = string("attn_1_cast_fp16")]; tensor var_162 = const()[name = string("op_162"), val = tensor([1, 1280, 1, -1])]; tensor input_1_cast_fp16 = reshape(shape = var_162, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143770368)))]; tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147047232)))]; tensor obj_11_cast_fp16 = conv(bias = layers_0_self_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_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; fp16 var_184_to_fp16 = const()[name = string("op_184_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_184_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147049856)))]; tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147052480)))]; fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147055104)))]; tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150331968)))]; 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_13_cast_fp16)[name = string("query_3_cast_fp16")]; tensor var_204 = const()[name = string("op_204"), val = tensor([1, 20, 64, -1])]; tensor mh_q_3_cast_fp16 = reshape(shape = var_204, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; fp16 var_206_to_fp16 = const()[name = string("op_206_to_fp16"), val = fp16(0x1p-3)]; tensor var_207_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_206_to_fp16)[name = string("op_207_cast_fp16")]; tensor var_208 = const()[name = string("op_208"), val = tensor([1, 20, 64, -1])]; tensor var_209_cast_fp16 = reshape(shape = var_208, x = obj_17_cast_fp16)[name = string("op_209_cast_fp16")]; bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(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_207_cast_fp16, y = var_209_cast_fp16)[name = string("mh_w_5_cast_fp16")]; tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; tensor var_213_axes_0 = const()[name = string("op_213_axes_0"), val = tensor([1])]; tensor var_213_cast_fp16 = expand_dims(axes = var_213_axes_0, x = read_state_4)[name = string("op_213_cast_fp16")]; tensor var_214_axes_0 = const()[name = string("op_214_axes_0"), val = tensor([2])]; tensor var_214_cast_fp16 = expand_dims(axes = var_214_axes_0, x = var_213_cast_fp16)[name = string("op_214_cast_fp16")]; tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_7_cast_fp16")]; tensor obj_23_cast_fp16 = softmax(axis = var_76, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; tensor var_218 = const()[name = string("op_218"), val = tensor([1, 20, 64, -1])]; tensor var_219_cast_fp16 = reshape(shape = var_218, x = obj_19_cast_fp16)[name = string("op_219_cast_fp16")]; bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_219_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; tensor var_222 = const()[name = string("op_222"), val = tensor([1, 1280, 1, -1])]; tensor input_3_cast_fp16 = reshape(shape = var_222, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150334592)))]; tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153611456)))]; tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_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_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; fp16 var_240_to_fp16 = const()[name = string("op_240_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_240_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153614080)))]; tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153616704)))]; fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(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_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153619328)))]; tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166726592)))]; 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 = string("input_7_cast_fp16")]; string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166736896)))]; tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179844160)))]; 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 = string("hidden_states_3_cast_fp16")]; tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 1280, 1, 1536])]; tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 1280, 1, 1536])]; tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; int32 var_285 = const()[name = string("op_285"), val = int32(3)]; tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; fp16 var_310_to_fp16 = const()[name = string("op_310_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_310_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179846784)))]; tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179849408)))]; fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179852032)))]; tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183128896)))]; 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_25_cast_fp16)[name = string("query_5_cast_fp16")]; string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)]; tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183131520)))]; 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_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)]; tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186408384)))]; tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189685248)))]; 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_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; tensor var_348_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_348_cast_fp16")]; tensor key_3_cast_fp16 = add(x = var_49_cast_fp16_1, y = var_348_cast_fp16)[name = string("key_3_cast_fp16")]; tensor var_350_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_350_cast_fp16")]; tensor value_3_cast_fp16 = add(x = var_56_cast_fp16_1, y = var_350_cast_fp16)[name = string("value_3_cast_fp16")]; tensor var_353 = const()[name = string("op_353"), val = tensor([1, 20, 64, -1])]; tensor mh_q_5_cast_fp16 = reshape(shape = var_353, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; fp16 var_355_to_fp16 = const()[name = string("op_355_to_fp16"), val = fp16(0x1p-3)]; tensor var_356_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_355_to_fp16)[name = string("op_356_cast_fp16")]; tensor var_357 = const()[name = string("op_357"), val = tensor([1, 20, 64, -1])]; tensor var_358_cast_fp16 = reshape(shape = var_357, x = key_3_cast_fp16)[name = string("op_358_cast_fp16")]; bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_356_cast_fp16, y = var_358_cast_fp16)[name = string("mh_w_9_cast_fp16")]; tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_11_cast_fp16")]; tensor var_366_cast_fp16 = softmax(axis = var_285, x = mh_w_11_cast_fp16)[name = string("op_366_cast_fp16")]; tensor var_367 = const()[name = string("op_367"), val = tensor([1, 20, 64, -1])]; tensor var_368_cast_fp16 = reshape(shape = var_367, x = value_3_cast_fp16)[name = string("op_368_cast_fp16")]; bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_368_cast_fp16, y = var_366_cast_fp16)[name = string("attn_5_cast_fp16")]; tensor var_371 = const()[name = string("op_371"), val = tensor([1, 1280, 1, -1])]; tensor input_11_cast_fp16 = reshape(shape = var_371, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189687872)))]; tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192964736)))]; tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; fp16 var_393_to_fp16 = const()[name = string("op_393_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_393_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192967360)))]; tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192969984)))]; fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192972608)))]; tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196249472)))]; 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_33_cast_fp16)[name = string("query_7_cast_fp16")]; tensor var_413 = const()[name = string("op_413"), val = tensor([1, 20, 64, -1])]; tensor mh_q_7_cast_fp16 = reshape(shape = var_413, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; fp16 var_415_to_fp16 = const()[name = string("op_415_to_fp16"), val = fp16(0x1p-3)]; tensor var_416_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_415_to_fp16)[name = string("op_416_cast_fp16")]; tensor var_417 = const()[name = string("op_417"), val = tensor([1, 20, 64, -1])]; tensor var_418_cast_fp16 = reshape(shape = var_417, x = obj_35_cast_fp16)[name = string("op_418_cast_fp16")]; bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(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_416_cast_fp16, y = var_418_cast_fp16)[name = string("mh_w_13_cast_fp16")]; tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_15_cast_fp16")]; tensor obj_41_cast_fp16 = softmax(axis = var_285, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; tensor var_427 = const()[name = string("op_427"), val = tensor([1, 20, 64, -1])]; tensor var_428_cast_fp16 = reshape(shape = var_427, x = obj_37_cast_fp16)[name = string("op_428_cast_fp16")]; bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_428_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; tensor var_431 = const()[name = string("op_431"), val = tensor([1, 1280, 1, -1])]; tensor input_13_cast_fp16 = reshape(shape = var_431, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")]; string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196252096)))]; tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199528960)))]; tensor obj_39_cast_fp16 = conv(bias = layers_1_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_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; fp16 var_449_to_fp16 = const()[name = string("op_449_to_fp16"), val = fp16(0x1.5p-17)]; tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_449_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199531584)))]; tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199534208)))]; fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(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_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199536832)))]; tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212644096)))]; 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 = string("input_17_cast_fp16")]; string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212654400)))]; tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225761664)))]; 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 = string("hidden_states_5_cast_fp16")]; tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 1280, 1, 1536])]; tensor