whisperkittools-0999a613c56c462b063b6b25d96260e1fc6ee2de generated files: dearyoungjo_whisper-tuned-dearyoungjo
4dd1b63
verified
| program(1.0) | |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.4.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] | |
| { | |
| func main<ios16>(tensor<int32, [1]> cache_length, tensor<fp16, [1, 448]> decoder_key_padding_mask, tensor<fp16, [1, 512, 1, 1500]> encoder_output_embeds, tensor<int32, [1]> input_ids, tensor<fp16, [1, 3072, 1, 448]> key_cache, tensor<fp16, [1, 448]> kv_cache_update_mask, tensor<fp16, [1, 3072, 1, 448]> value_cache) { | |
| tensor<int32, []> var_28_axis_0 = const()[name = tensor<string, []>("op_28_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> var_28_batch_dims_0 = const()[name = tensor<string, []>("op_28_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [51864, 512]> embed_tokens_weight_to_fp16 = const()[name = tensor<string, []>("embed_tokens_weight_to_fp16"), val = tensor<fp16, [51864, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<fp16, [1, 512]> var_28_cast_fp16 = gather(axis = var_28_axis_0, batch_dims = var_28_batch_dims_0, indices = input_ids, x = embed_tokens_weight_to_fp16)[name = tensor<string, []>("op_28_cast_fp16")]; | |
| tensor<int32, []> var_32_axis_0 = const()[name = tensor<string, []>("op_32_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> var_32_batch_dims_0 = const()[name = tensor<string, []>("op_32_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [448, 512]> embed_positions_weight_to_fp16 = const()[name = tensor<string, []>("embed_positions_weight_to_fp16"), val = tensor<fp16, [448, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53108864)))]; | |
| tensor<fp16, [1, 512]> var_32_cast_fp16 = gather(axis = var_32_axis_0, batch_dims = var_32_batch_dims_0, indices = cache_length, x = embed_positions_weight_to_fp16)[name = tensor<string, []>("op_32_cast_fp16")]; | |
| tensor<fp16, [1, 512]> hidden_states_1_cast_fp16 = add(x = var_28_cast_fp16, y = var_32_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")]; | |
| tensor<int32, [1]> var_46_axes_0 = const()[name = tensor<string, []>("op_46_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 512, 1]> var_46_cast_fp16 = expand_dims(axes = var_46_axes_0, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("op_46_cast_fp16")]; | |
| tensor<int32, [1]> inputs_1_axes_0 = const()[name = tensor<string, []>("inputs_1_axes_0"), val = tensor<int32, [1]>([3])]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_46_cast_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")]; | |
| tensor<int32, [6]> tile_0 = const()[name = tensor<string, []>("tile_0"), val = tensor<int32, [6]>([512, 512, 512, 512, 512, 512])]; | |
| tensor<int32, []> var_51_axis_0 = const()[name = tensor<string, []>("op_51_axis_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_0, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_1, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_2, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_3, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_4, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_5 = split(axis = var_51_axis_0, split_sizes = tile_0, x = key_cache)[name = tensor<string, []>("op_51_cast_fp16")]; | |
| tensor<int32, [6]> tile_1 = const()[name = tensor<string, []>("tile_1"), val = tensor<int32, [6]>([512, 512, 512, 512, 512, 512])]; | |
| tensor<int32, []> var_60_axis_0 = const()[name = tensor<string, []>("op_60_axis_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_0, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_1, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_2, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_3, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_4, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_5 = split(axis = var_60_axis_0, split_sizes = tile_1, x = value_cache)[name = tensor<string, []>("op_60_cast_fp16")]; | |
| tensor<int32, []> var_72 = const()[name = tensor<string, []>("op_72"), val = tensor<int32, []>(3)]; | |
| tensor<int32, [1]> out_1_axes_0 = const()[name = tensor<string, []>("out_1_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_98_to_fp16 = const()[name = tensor<string, []>("op_98_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_98_to_fp16, x = inputs_1_cast_fp16)[name = tensor<string, []>("out_1_cast_fp16")]; | |
| tensor<fp16, [512]> obj_1_mean_0_to_fp16 = const()[name = tensor<string, []>("obj_1_mean_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53567680)))]; | |
| tensor<fp16, [512]> obj_1_variance_0_to_fp16 = const()[name = tensor<string, []>("obj_1_variance_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53568768)))]; | |
| tensor<fp16, [512]> obj_1_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_1_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53569856)))]; | |
| tensor<fp16, [512]> obj_1_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_1_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53570944)))]; | |
| tensor<fp16, []> obj_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_1_cast_fp16")]; | |
| tensor<string, []> query_1_pad_type_0 = const()[name = tensor<string, []>("query_1_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_1_strides_0 = const()[name = tensor<string, []>("query_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_1_pad_0 = const()[name = tensor<string, []>("query_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_1_dilations_0 = const()[name = tensor<string, []>("query_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_1_groups_0 = const()[name = tensor<string, []>("query_1_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53572032)))]; | |
| tensor<fp16, [512]> layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54096384)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_1_cast_fp16")]; | |
| tensor<string, []> current_key_1_pad_type_0 = const()[name = tensor<string, []>("current_key_1_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_key_1_strides_0 = const()[name = tensor<string, []>("current_key_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_key_1_pad_0 = const()[name = tensor<string, []>("current_key_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_key_1_dilations_0 = const()[name = tensor<string, []>("current_key_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_key_1_groups_0 = const()[name = tensor<string, []>("current_key_1_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54097472)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_key_1_cast_fp16")]; | |
| tensor<string, []> current_value_1_pad_type_0 = const()[name = tensor<string, []>("current_value_1_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_value_1_strides_0 = const()[name = tensor<string, []>("current_value_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_value_1_pad_0 = const()[name = tensor<string, []>("current_value_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_value_1_dilations_0 = const()[name = tensor<string, []>("current_value_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_value_1_groups_0 = const()[name = tensor<string, []>("current_value_1_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54621824)))]; | |
| tensor<fp16, [512]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55146176)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_value_1_cast_fp16")]; | |
| tensor<int32, [1]> var_133_axes_0 = const()[name = tensor<string, []>("op_133_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, [1, 1, 448]> var_133_cast_fp16 = expand_dims(axes = var_133_axes_0, x = kv_cache_update_mask)[name = tensor<string, []>("op_133_cast_fp16")]; | |
| tensor<int32, [1]> var_134_axes_0 = const()[name = tensor<string, []>("op_134_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 1, 1, 448]> var_134_cast_fp16 = expand_dims(axes = var_134_axes_0, x = var_133_cast_fp16)[name = tensor<string, []>("op_134_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_136_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_136_cast_fp16")]; | |
| tensor<fp16, []> var_73_to_fp16 = const()[name = tensor<string, []>("op_73_to_fp16"), val = tensor<fp16, []>(0x1p+0)]; | |
| tensor<fp16, [1, 1, 1, 448]> var_137_cast_fp16 = sub(x = var_73_to_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_137_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_138_cast_fp16 = mul(x = var_51_cast_fp16_0, y = var_137_cast_fp16)[name = tensor<string, []>("op_138_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_1_cast_fp16 = add(x = var_136_cast_fp16, y = var_138_cast_fp16)[name = tensor<string, []>("key_1_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_140_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_140_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_142_cast_fp16 = mul(x = var_60_cast_fp16_0, y = var_137_cast_fp16)[name = tensor<string, []>("op_142_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_1_cast_fp16 = add(x = var_140_cast_fp16, y = var_142_cast_fp16)[name = tensor<string, []>("value_1_cast_fp16")]; | |
| tensor<int32, [4]> var_145 = const()[name = tensor<string, []>("op_145"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_1_cast_fp16 = reshape(shape = var_145, x = query_1_cast_fp16)[name = tensor<string, []>("mh_q_1_cast_fp16")]; | |
| tensor<fp16, []> var_147_to_fp16 = const()[name = tensor<string, []>("op_147_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_148_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_147_to_fp16)[name = tensor<string, []>("op_148_cast_fp16")]; | |
| tensor<int32, [4]> var_149 = const()[name = tensor<string, []>("op_149"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_150_cast_fp16 = reshape(shape = var_149, x = key_1_cast_fp16)[name = tensor<string, []>("op_150_cast_fp16")]; | |
| tensor<bool, []> mh_w_1_transpose_x_0 = const()[name = tensor<string, []>("mh_w_1_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_1_transpose_y_0 = const()[name = tensor<string, []>("mh_w_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_148_cast_fp16, y = var_150_cast_fp16)[name = tensor<string, []>("mh_w_1_cast_fp16")]; | |
| tensor<int32, [1]> var_154_axes_0 = const()[name = tensor<string, []>("op_154_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, [1, 1, 448]> var_154_cast_fp16 = expand_dims(axes = var_154_axes_0, x = decoder_key_padding_mask)[name = tensor<string, []>("op_154_cast_fp16")]; | |
| tensor<int32, [1]> var_155_axes_0 = const()[name = tensor<string, []>("op_155_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 1, 1, 448]> var_155_cast_fp16 = expand_dims(axes = var_155_axes_0, x = var_154_cast_fp16)[name = tensor<string, []>("op_155_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_155_cast_fp16)[name = tensor<string, []>("mh_w_3_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_158_cast_fp16 = softmax(axis = var_72, x = mh_w_3_cast_fp16)[name = tensor<string, []>("op_158_cast_fp16")]; | |
| tensor<int32, [4]> var_159 = const()[name = tensor<string, []>("op_159"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_160_cast_fp16 = reshape(shape = var_159, x = value_1_cast_fp16)[name = tensor<string, []>("op_160_cast_fp16")]; | |
| tensor<bool, []> attn_1_transpose_x_0 = const()[name = tensor<string, []>("attn_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_1_transpose_y_0 = const()[name = tensor<string, []>("attn_1_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_160_cast_fp16, y = var_158_cast_fp16)[name = tensor<string, []>("attn_1_cast_fp16")]; | |
| tensor<int32, [4]> var_163 = const()[name = tensor<string, []>("op_163"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_1_cast_fp16 = reshape(shape = var_163, x = attn_1_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")]; | |
| tensor<string, []> obj_7_pad_type_0 = const()[name = tensor<string, []>("obj_7_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_7_strides_0 = const()[name = tensor<string, []>("obj_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_7_pad_0 = const()[name = tensor<string, []>("obj_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_7_dilations_0 = const()[name = tensor<string, []>("obj_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_7_groups_0 = const()[name = tensor<string, []>("obj_7_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55147264)))]; | |
| tensor<fp16, [512]> layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55671616)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_7_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_7_cast_fp16)[name = tensor<string, []>("inputs_3_cast_fp16")]; | |
| tensor<int32, [1]> out_3_axes_0 = const()[name = tensor<string, []>("out_3_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_185_to_fp16 = const()[name = tensor<string, []>("op_185_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_185_to_fp16, x = inputs_3_cast_fp16)[name = tensor<string, []>("out_3_cast_fp16")]; | |
| tensor<fp16, [512]> obj_9_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_9_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55672704)))]; | |
| tensor<fp16, [512]> obj_9_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_9_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55673792)))]; | |
| tensor<fp16, []> obj_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_9_cast_fp16")]; | |
| tensor<string, []> query_3_pad_type_0 = const()[name = tensor<string, []>("query_3_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_3_strides_0 = const()[name = tensor<string, []>("query_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_3_pad_0 = const()[name = tensor<string, []>("query_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_3_dilations_0 = const()[name = tensor<string, []>("query_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_3_groups_0 = const()[name = tensor<string, []>("query_3_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55674880)))]; | |
| tensor<fp16, [512]> layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56199232)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_3_cast_fp16")]; | |
| tensor<string, []> key_3_pad_type_0 = const()[name = tensor<string, []>("key_3_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> key_3_strides_0 = const()[name = tensor<string, []>("key_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> key_3_pad_0 = const()[name = tensor<string, []>("key_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> key_3_dilations_0 = const()[name = tensor<string, []>("key_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> key_3_groups_0 = const()[name = tensor<string, []>("key_3_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56200320)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("key_3_cast_fp16")]; | |
| tensor<string, []> value_3_pad_type_0 = const()[name = tensor<string, []>("value_3_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> value_3_strides_0 = const()[name = tensor<string, []>("value_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> value_3_pad_0 = const()[name = tensor<string, []>("value_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> value_3_dilations_0 = const()[name = tensor<string, []>("value_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> value_3_groups_0 = const()[name = tensor<string, []>("value_3_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56724672)))]; | |
| tensor<fp16, [512]> layers_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57249024)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("value_3_cast_fp16")]; | |
| tensor<int32, [4]> var_220 = const()[name = tensor<string, []>("op_220"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_3_cast_fp16 = reshape(shape = var_220, x = query_3_cast_fp16)[name = tensor<string, []>("mh_q_3_cast_fp16")]; | |
| tensor<fp16, []> var_222_to_fp16 = const()[name = tensor<string, []>("op_222_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_223_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_222_to_fp16)[name = tensor<string, []>("op_223_cast_fp16")]; | |
| tensor<int32, [4]> var_224 = const()[name = tensor<string, []>("op_224"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_225_cast_fp16 = reshape(shape = var_224, x = key_3_cast_fp16)[name = tensor<string, []>("op_225_cast_fp16")]; | |
| tensor<bool, []> mh_w_5_transpose_x_0 = const()[name = tensor<string, []>("mh_w_5_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_5_transpose_y_0 = const()[name = tensor<string, []>("mh_w_5_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 1500]> mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_223_cast_fp16, y = var_225_cast_fp16)[name = tensor<string, []>("mh_w_5_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_13_cast_fp16 = softmax(axis = var_72, x = mh_w_5_cast_fp16)[name = tensor<string, []>("obj_13_cast_fp16")]; | |
| tensor<int32, [4]> var_229 = const()[name = tensor<string, []>("op_229"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_230_cast_fp16 = reshape(shape = var_229, x = value_3_cast_fp16)[name = tensor<string, []>("op_230_cast_fp16")]; | |
| tensor<bool, []> attn_3_transpose_x_0 = const()[name = tensor<string, []>("attn_3_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_3_transpose_y_0 = const()[name = tensor<string, []>("attn_3_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_230_cast_fp16, y = obj_13_cast_fp16)[name = tensor<string, []>("attn_3_cast_fp16")]; | |
| tensor<int32, [4]> var_233 = const()[name = tensor<string, []>("op_233"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_3_cast_fp16 = reshape(shape = var_233, x = attn_3_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")]; | |
| tensor<string, []> obj_11_pad_type_0 = const()[name = tensor<string, []>("obj_11_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_11_strides_0 = const()[name = tensor<string, []>("obj_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_11_pad_0 = const()[name = tensor<string, []>("obj_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_11_dilations_0 = const()[name = tensor<string, []>("obj_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_11_groups_0 = const()[name = tensor<string, []>("obj_11_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57250112)))]; | |
| tensor<fp16, [512]> layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57774464)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_11_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_11_cast_fp16)[name = tensor<string, []>("inputs_5_cast_fp16")]; | |
| tensor<int32, [1]> out_5_axes_0 = const()[name = tensor<string, []>("out_5_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_251_to_fp16 = const()[name = tensor<string, []>("op_251_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_251_to_fp16, x = inputs_5_cast_fp16)[name = tensor<string, []>("out_5_cast_fp16")]; | |
| tensor<fp16, [512]> input_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_5_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57775552)))]; | |
| tensor<fp16, [512]> input_5_beta_0_to_fp16 = const()[name = tensor<string, []>("input_5_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57776640)))]; | |
| tensor<fp16, []> input_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("input_5_cast_fp16")]; | |
| tensor<string, []> input_7_pad_type_0 = const()[name = tensor<string, []>("input_7_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> input_7_strides_0 = const()[name = tensor<string, []>("input_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> input_7_pad_0 = const()[name = tensor<string, []>("input_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> input_7_dilations_0 = const()[name = tensor<string, []>("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> input_7_groups_0 = const()[name = tensor<string, []>("input_7_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [2048, 512, 1, 1]> layers_0_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57777728)))]; | |
| tensor<fp16, [2048]> layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59874944)))]; | |
| tensor<fp16, [1, 2048, 1, 1]> 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<string, []>("input_7_cast_fp16")]; | |
| tensor<string, []> input_9_mode_0 = const()[name = tensor<string, []>("input_9_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 2048, 1, 1]> input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")]; | |
| tensor<string, []> hidden_states_3_pad_type_0 = const()[name = tensor<string, []>("hidden_states_3_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> hidden_states_3_strides_0 = const()[name = tensor<string, []>("hidden_states_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = tensor<string, []>("hidden_states_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> hidden_states_3_dilations_0 = const()[name = tensor<string, []>("hidden_states_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> hidden_states_3_groups_0 = const()[name = tensor<string, []>("hidden_states_3_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 2048, 1, 1]> layers_0_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59879104)))]; | |
| tensor<fp16, [512]> layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61976320)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("hidden_states_3_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = tensor<string, []>("inputs_7_cast_fp16")]; | |
| tensor<int32, []> var_286 = const()[name = tensor<string, []>("op_286"), val = tensor<int32, []>(3)]; | |
| tensor<int32, [1]> out_7_axes_0 = const()[name = tensor<string, []>("out_7_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_312_to_fp16 = const()[name = tensor<string, []>("op_312_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_312_to_fp16, x = inputs_7_cast_fp16)[name = tensor<string, []>("out_7_cast_fp16")]; | |
| tensor<fp16, [512]> obj_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_15_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61977408)))]; | |
| tensor<fp16, [512]> obj_15_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_15_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61978496)))]; | |
| tensor<fp16, []> obj_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_15_cast_fp16")]; | |
| tensor<string, []> query_5_pad_type_0 = const()[name = tensor<string, []>("query_5_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_5_strides_0 = const()[name = tensor<string, []>("query_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_5_pad_0 = const()[name = tensor<string, []>("query_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_5_dilations_0 = const()[name = tensor<string, []>("query_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_5_groups_0 = const()[name = tensor<string, []>("query_5_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61979584)))]; | |
| tensor<fp16, [512]> layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62503936)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_5_cast_fp16")]; | |
| tensor<string, []> current_key_3_pad_type_0 = const()[name = tensor<string, []>("current_key_3_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_key_3_strides_0 = const()[name = tensor<string, []>("current_key_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_key_3_pad_0 = const()[name = tensor<string, []>("current_key_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_key_3_dilations_0 = const()[name = tensor<string, []>("current_key_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_key_3_groups_0 = const()[name = tensor<string, []>("current_key_3_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62505024)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_key_3_cast_fp16")]; | |
| tensor<string, []> current_value_3_pad_type_0 = const()[name = tensor<string, []>("current_value_3_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_value_3_strides_0 = const()[name = tensor<string, []>("current_value_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_value_3_pad_0 = const()[name = tensor<string, []>("current_value_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_value_3_dilations_0 = const()[name = tensor<string, []>("current_value_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_value_3_groups_0 = const()[name = tensor<string, []>("current_value_3_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63029376)))]; | |
| tensor<fp16, [512]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63553728)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_value_3_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_350_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_350_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_352_cast_fp16 = mul(x = var_51_cast_fp16_1, y = var_137_cast_fp16)[name = tensor<string, []>("op_352_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_5_cast_fp16 = add(x = var_350_cast_fp16, y = var_352_cast_fp16)[name = tensor<string, []>("key_5_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_354_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_354_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_356_cast_fp16 = mul(x = var_60_cast_fp16_1, y = var_137_cast_fp16)[name = tensor<string, []>("op_356_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_5_cast_fp16 = add(x = var_354_cast_fp16, y = var_356_cast_fp16)[name = tensor<string, []>("value_5_cast_fp16")]; | |
| tensor<int32, [4]> var_359 = const()[name = tensor<string, []>("op_359"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_5_cast_fp16 = reshape(shape = var_359, x = query_5_cast_fp16)[name = tensor<string, []>("mh_q_5_cast_fp16")]; | |
| tensor<fp16, []> var_361_to_fp16 = const()[name = tensor<string, []>("op_361_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_362_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_361_to_fp16)[name = tensor<string, []>("op_362_cast_fp16")]; | |
| tensor<int32, [4]> var_363 = const()[name = tensor<string, []>("op_363"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_364_cast_fp16 = reshape(shape = var_363, x = key_5_cast_fp16)[name = tensor<string, []>("op_364_cast_fp16")]; | |
| tensor<bool, []> mh_w_7_transpose_x_0 = const()[name = tensor<string, []>("mh_w_7_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_7_transpose_y_0 = const()[name = tensor<string, []>("mh_w_7_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_362_cast_fp16, y = var_364_cast_fp16)[name = tensor<string, []>("mh_w_7_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_9_cast_fp16 = add(x = mh_w_7_cast_fp16, y = var_155_cast_fp16)[name = tensor<string, []>("mh_w_9_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_372_cast_fp16 = softmax(axis = var_286, x = mh_w_9_cast_fp16)[name = tensor<string, []>("op_372_cast_fp16")]; | |
| tensor<int32, [4]> var_373 = const()[name = tensor<string, []>("op_373"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_374_cast_fp16 = reshape(shape = var_373, x = value_5_cast_fp16)[name = tensor<string, []>("op_374_cast_fp16")]; | |
| tensor<bool, []> attn_5_transpose_x_0 = const()[name = tensor<string, []>("attn_5_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_5_transpose_y_0 = const()[name = tensor<string, []>("attn_5_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_374_cast_fp16, y = var_372_cast_fp16)[name = tensor<string, []>("attn_5_cast_fp16")]; | |
| tensor<int32, [4]> var_377 = const()[name = tensor<string, []>("op_377"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_11_cast_fp16 = reshape(shape = var_377, x = attn_5_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; | |
| tensor<string, []> obj_21_pad_type_0 = const()[name = tensor<string, []>("obj_21_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_21_strides_0 = const()[name = tensor<string, []>("obj_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_21_pad_0 = const()[name = tensor<string, []>("obj_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_21_dilations_0 = const()[name = tensor<string, []>("obj_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_21_groups_0 = const()[name = tensor<string, []>("obj_21_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63554816)))]; | |
| tensor<fp16, [512]> layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64079168)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_21_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_21_cast_fp16)[name = tensor<string, []>("inputs_9_cast_fp16")]; | |
| tensor<int32, [1]> out_9_axes_0 = const()[name = tensor<string, []>("out_9_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_399_to_fp16 = const()[name = tensor<string, []>("op_399_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_399_to_fp16, x = inputs_9_cast_fp16)[name = tensor<string, []>("out_9_cast_fp16")]; | |
| tensor<fp16, [512]> obj_23_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_23_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64080256)))]; | |
| tensor<fp16, [512]> obj_23_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_23_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64081344)))]; | |
| tensor<fp16, []> obj_23_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_23_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_23_cast_fp16")]; | |
| tensor<string, []> query_7_pad_type_0 = const()[name = tensor<string, []>("query_7_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_7_strides_0 = const()[name = tensor<string, []>("query_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_7_pad_0 = const()[name = tensor<string, []>("query_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_7_dilations_0 = const()[name = tensor<string, []>("query_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_7_groups_0 = const()[name = tensor<string, []>("query_7_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64082432)))]; | |
| tensor<fp16, [512]> layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64606784)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_7_cast_fp16")]; | |
| tensor<string, []> key_7_pad_type_0 = const()[name = tensor<string, []>("key_7_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> key_7_strides_0 = const()[name = tensor<string, []>("key_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> key_7_pad_0 = const()[name = tensor<string, []>("key_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> key_7_dilations_0 = const()[name = tensor<string, []>("key_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> key_7_groups_0 = const()[name = tensor<string, []>("key_7_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64607872)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("key_7_cast_fp16")]; | |
| tensor<string, []> value_7_pad_type_0 = const()[name = tensor<string, []>("value_7_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> value_7_strides_0 = const()[name = tensor<string, []>("value_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> value_7_pad_0 = const()[name = tensor<string, []>("value_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> value_7_dilations_0 = const()[name = tensor<string, []>("value_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> value_7_groups_0 = const()[name = tensor<string, []>("value_7_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65132224)))]; | |
| tensor<fp16, [512]> layers_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65656576)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("value_7_cast_fp16")]; | |
| tensor<int32, [4]> var_434 = const()[name = tensor<string, []>("op_434"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_7_cast_fp16 = reshape(shape = var_434, x = query_7_cast_fp16)[name = tensor<string, []>("mh_q_7_cast_fp16")]; | |
| tensor<fp16, []> var_436_to_fp16 = const()[name = tensor<string, []>("op_436_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_437_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_436_to_fp16)[name = tensor<string, []>("op_437_cast_fp16")]; | |
| tensor<int32, [4]> var_438 = const()[name = tensor<string, []>("op_438"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_439_cast_fp16 = reshape(shape = var_438, x = key_7_cast_fp16)[name = tensor<string, []>("op_439_cast_fp16")]; | |
| tensor<bool, []> mh_w_11_transpose_x_0 = const()[name = tensor<string, []>("mh_w_11_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_11_transpose_y_0 = const()[name = tensor<string, []>("mh_w_11_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 1500]> mh_w_11_cast_fp16 = matmul(transpose_x = mh_w_11_transpose_x_0, transpose_y = mh_w_11_transpose_y_0, x = var_437_cast_fp16, y = var_439_cast_fp16)[name = tensor<string, []>("mh_w_11_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_27_cast_fp16 = softmax(axis = var_286, x = mh_w_11_cast_fp16)[name = tensor<string, []>("obj_27_cast_fp16")]; | |
| tensor<int32, [4]> var_443 = const()[name = tensor<string, []>("op_443"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_444_cast_fp16 = reshape(shape = var_443, x = value_7_cast_fp16)[name = tensor<string, []>("op_444_cast_fp16")]; | |
| tensor<bool, []> attn_7_transpose_x_0 = const()[name = tensor<string, []>("attn_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_7_transpose_y_0 = const()[name = tensor<string, []>("attn_7_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_444_cast_fp16, y = obj_27_cast_fp16)[name = tensor<string, []>("attn_7_cast_fp16")]; | |
| tensor<int32, [4]> var_447 = const()[name = tensor<string, []>("op_447"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_13_cast_fp16 = reshape(shape = var_447, x = attn_7_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; | |
| tensor<string, []> obj_25_pad_type_0 = const()[name = tensor<string, []>("obj_25_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_25_strides_0 = const()[name = tensor<string, []>("obj_25_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_25_pad_0 = const()[name = tensor<string, []>("obj_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_25_dilations_0 = const()[name = tensor<string, []>("obj_25_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_25_groups_0 = const()[name = tensor<string, []>("obj_25_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65657664)))]; | |
| tensor<fp16, [512]> layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66182016)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_25_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_25_cast_fp16)[name = tensor<string, []>("inputs_11_cast_fp16")]; | |
| tensor<int32, [1]> out_11_axes_0 = const()[name = tensor<string, []>("out_11_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_465_to_fp16 = const()[name = tensor<string, []>("op_465_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_465_to_fp16, x = inputs_11_cast_fp16)[name = tensor<string, []>("out_11_cast_fp16")]; | |
| tensor<fp16, [512]> input_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_15_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66183104)))]; | |
| tensor<fp16, [512]> input_15_beta_0_to_fp16 = const()[name = tensor<string, []>("input_15_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66184192)))]; | |
| tensor<fp16, []> input_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("input_15_cast_fp16")]; | |
| tensor<string, []> input_17_pad_type_0 = const()[name = tensor<string, []>("input_17_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> input_17_strides_0 = const()[name = tensor<string, []>("input_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> input_17_pad_0 = const()[name = tensor<string, []>("input_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> input_17_dilations_0 = const()[name = tensor<string, []>("input_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> input_17_groups_0 = const()[name = tensor<string, []>("input_17_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [2048, 512, 1, 1]> layers_1_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66185280)))]; | |
| tensor<fp16, [2048]> layers_1_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68282496)))]; | |
| tensor<fp16, [1, 2048, 1, 1]> 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<string, []>("input_17_cast_fp16")]; | |
| tensor<string, []> input_19_mode_0 = const()[name = tensor<string, []>("input_19_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 2048, 1, 1]> input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; | |
| tensor<string, []> hidden_states_5_pad_type_0 = const()[name = tensor<string, []>("hidden_states_5_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> hidden_states_5_strides_0 = const()[name = tensor<string, []>("hidden_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> hidden_states_5_pad_0 = const()[name = tensor<string, []>("hidden_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> hidden_states_5_dilations_0 = const()[name = tensor<string, []>("hidden_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> hidden_states_5_groups_0 = const()[name = tensor<string, []>("hidden_states_5_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 2048, 1, 1]> layers_1_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68286656)))]; | |
| tensor<fp16, [512]> layers_1_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70383872)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("hidden_states_5_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor<string, []>("inputs_13_cast_fp16")]; | |
| tensor<int32, []> var_500 = const()[name = tensor<string, []>("op_500"), val = tensor<int32, []>(3)]; | |
| tensor<int32, [1]> out_13_axes_0 = const()[name = tensor<string, []>("out_13_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_526_to_fp16 = const()[name = tensor<string, []>("op_526_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_526_to_fp16, x = inputs_13_cast_fp16)[name = tensor<string, []>("out_13_cast_fp16")]; | |
| tensor<fp16, [512]> obj_29_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_29_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70384960)))]; | |
| tensor<fp16, [512]> obj_29_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_29_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70386048)))]; | |
| tensor<fp16, []> obj_29_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_29_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_29_cast_fp16")]; | |
| tensor<string, []> query_9_pad_type_0 = const()[name = tensor<string, []>("query_9_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_9_strides_0 = const()[name = tensor<string, []>("query_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_9_pad_0 = const()[name = tensor<string, []>("query_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_9_dilations_0 = const()[name = tensor<string, []>("query_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_9_groups_0 = const()[name = tensor<string, []>("query_9_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70387136)))]; | |
| tensor<fp16, [512]> layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70911488)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_9_cast_fp16")]; | |
| tensor<string, []> current_key_5_pad_type_0 = const()[name = tensor<string, []>("current_key_5_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_key_5_strides_0 = const()[name = tensor<string, []>("current_key_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_key_5_pad_0 = const()[name = tensor<string, []>("current_key_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_key_5_dilations_0 = const()[name = tensor<string, []>("current_key_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_key_5_groups_0 = const()[name = tensor<string, []>("current_key_5_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70912576)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_key_5_cast_fp16")]; | |
| tensor<string, []> current_value_5_pad_type_0 = const()[name = tensor<string, []>("current_value_5_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_value_5_strides_0 = const()[name = tensor<string, []>("current_value_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_value_5_pad_0 = const()[name = tensor<string, []>("current_value_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_value_5_dilations_0 = const()[name = tensor<string, []>("current_value_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_value_5_groups_0 = const()[name = tensor<string, []>("current_value_5_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71436928)))]; | |
| tensor<fp16, [512]> layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71961280)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_value_5_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_564_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_564_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_566_cast_fp16 = mul(x = var_51_cast_fp16_2, y = var_137_cast_fp16)[name = tensor<string, []>("op_566_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_9_cast_fp16 = add(x = var_564_cast_fp16, y = var_566_cast_fp16)[name = tensor<string, []>("key_9_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_568_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_568_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_570_cast_fp16 = mul(x = var_60_cast_fp16_2, y = var_137_cast_fp16)[name = tensor<string, []>("op_570_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_9_cast_fp16 = add(x = var_568_cast_fp16, y = var_570_cast_fp16)[name = tensor<string, []>("value_9_cast_fp16")]; | |
| tensor<int32, [4]> var_573 = const()[name = tensor<string, []>("op_573"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_9_cast_fp16 = reshape(shape = var_573, x = query_9_cast_fp16)[name = tensor<string, []>("mh_q_9_cast_fp16")]; | |
| tensor<fp16, []> var_575_to_fp16 = const()[name = tensor<string, []>("op_575_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_576_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_575_to_fp16)[name = tensor<string, []>("op_576_cast_fp16")]; | |
| tensor<int32, [4]> var_577 = const()[name = tensor<string, []>("op_577"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_578_cast_fp16 = reshape(shape = var_577, x = key_9_cast_fp16)[name = tensor<string, []>("op_578_cast_fp16")]; | |
| tensor<bool, []> mh_w_13_transpose_x_0 = const()[name = tensor<string, []>("mh_w_13_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_13_transpose_y_0 = const()[name = tensor<string, []>("mh_w_13_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_576_cast_fp16, y = var_578_cast_fp16)[name = tensor<string, []>("mh_w_13_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_155_cast_fp16)[name = tensor<string, []>("mh_w_15_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_586_cast_fp16 = softmax(axis = var_500, x = mh_w_15_cast_fp16)[name = tensor<string, []>("op_586_cast_fp16")]; | |
| tensor<int32, [4]> var_587 = const()[name = tensor<string, []>("op_587"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_588_cast_fp16 = reshape(shape = var_587, x = value_9_cast_fp16)[name = tensor<string, []>("op_588_cast_fp16")]; | |
| tensor<bool, []> attn_9_transpose_x_0 = const()[name = tensor<string, []>("attn_9_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_9_transpose_y_0 = const()[name = tensor<string, []>("attn_9_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_588_cast_fp16, y = var_586_cast_fp16)[name = tensor<string, []>("attn_9_cast_fp16")]; | |
| tensor<int32, [4]> var_591 = const()[name = tensor<string, []>("op_591"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_21_cast_fp16 = reshape(shape = var_591, x = attn_9_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; | |
| tensor<string, []> obj_35_pad_type_0 = const()[name = tensor<string, []>("obj_35_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_35_strides_0 = const()[name = tensor<string, []>("obj_35_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_35_pad_0 = const()[name = tensor<string, []>("obj_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_35_dilations_0 = const()[name = tensor<string, []>("obj_35_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_35_groups_0 = const()[name = tensor<string, []>("obj_35_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71962368)))]; | |
| tensor<fp16, [512]> layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72486720)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_35_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_35_cast_fp16)[name = tensor<string, []>("inputs_15_cast_fp16")]; | |
| tensor<int32, [1]> out_15_axes_0 = const()[name = tensor<string, []>("out_15_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_613_to_fp16 = const()[name = tensor<string, []>("op_613_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_613_to_fp16, x = inputs_15_cast_fp16)[name = tensor<string, []>("out_15_cast_fp16")]; | |
| tensor<fp16, [512]> obj_37_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_37_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72487808)))]; | |
| tensor<fp16, [512]> obj_37_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_37_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72488896)))]; | |
| tensor<fp16, []> obj_37_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_37_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_37_cast_fp16")]; | |
| tensor<string, []> query_11_pad_type_0 = const()[name = tensor<string, []>("query_11_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_11_strides_0 = const()[name = tensor<string, []>("query_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_11_pad_0 = const()[name = tensor<string, []>("query_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_11_dilations_0 = const()[name = tensor<string, []>("query_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_11_groups_0 = const()[name = tensor<string, []>("query_11_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72489984)))]; | |
| tensor<fp16, [512]> layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73014336)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_11_cast_fp16")]; | |
| tensor<string, []> key_11_pad_type_0 = const()[name = tensor<string, []>("key_11_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> key_11_strides_0 = const()[name = tensor<string, []>("key_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> key_11_pad_0 = const()[name = tensor<string, []>("key_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> key_11_dilations_0 = const()[name = tensor<string, []>("key_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> key_11_groups_0 = const()[name = tensor<string, []>("key_11_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73015424)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("key_11_cast_fp16")]; | |
| tensor<string, []> value_11_pad_type_0 = const()[name = tensor<string, []>("value_11_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> value_11_strides_0 = const()[name = tensor<string, []>("value_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> value_11_pad_0 = const()[name = tensor<string, []>("value_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> value_11_dilations_0 = const()[name = tensor<string, []>("value_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> value_11_groups_0 = const()[name = tensor<string, []>("value_11_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73539776)))]; | |
| tensor<fp16, [512]> layers_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74064128)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("value_11_cast_fp16")]; | |
| tensor<int32, [4]> var_648 = const()[name = tensor<string, []>("op_648"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_11_cast_fp16 = reshape(shape = var_648, x = query_11_cast_fp16)[name = tensor<string, []>("mh_q_11_cast_fp16")]; | |
| tensor<fp16, []> var_650_to_fp16 = const()[name = tensor<string, []>("op_650_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_651_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_650_to_fp16)[name = tensor<string, []>("op_651_cast_fp16")]; | |
| tensor<int32, [4]> var_652 = const()[name = tensor<string, []>("op_652"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_653_cast_fp16 = reshape(shape = var_652, x = key_11_cast_fp16)[name = tensor<string, []>("op_653_cast_fp16")]; | |
| tensor<bool, []> mh_w_17_transpose_x_0 = const()[name = tensor<string, []>("mh_w_17_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_17_transpose_y_0 = const()[name = tensor<string, []>("mh_w_17_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 1500]> mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_651_cast_fp16, y = var_653_cast_fp16)[name = tensor<string, []>("mh_w_17_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_41_cast_fp16 = softmax(axis = var_500, x = mh_w_17_cast_fp16)[name = tensor<string, []>("obj_41_cast_fp16")]; | |
| tensor<int32, [4]> var_657 = const()[name = tensor<string, []>("op_657"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_658_cast_fp16 = reshape(shape = var_657, x = value_11_cast_fp16)[name = tensor<string, []>("op_658_cast_fp16")]; | |
| tensor<bool, []> attn_11_transpose_x_0 = const()[name = tensor<string, []>("attn_11_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_11_transpose_y_0 = const()[name = tensor<string, []>("attn_11_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_658_cast_fp16, y = obj_41_cast_fp16)[name = tensor<string, []>("attn_11_cast_fp16")]; | |
| tensor<int32, [4]> var_661 = const()[name = tensor<string, []>("op_661"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_23_cast_fp16 = reshape(shape = var_661, x = attn_11_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; | |
| tensor<string, []> obj_39_pad_type_0 = const()[name = tensor<string, []>("obj_39_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_39_strides_0 = const()[name = tensor<string, []>("obj_39_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_39_pad_0 = const()[name = tensor<string, []>("obj_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_39_dilations_0 = const()[name = tensor<string, []>("obj_39_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_39_groups_0 = const()[name = tensor<string, []>("obj_39_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74065216)))]; | |
| tensor<fp16, [512]> layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74589568)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_39_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_39_cast_fp16)[name = tensor<string, []>("inputs_17_cast_fp16")]; | |
| tensor<int32, [1]> out_17_axes_0 = const()[name = tensor<string, []>("out_17_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_679_to_fp16 = const()[name = tensor<string, []>("op_679_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_679_to_fp16, x = inputs_17_cast_fp16)[name = tensor<string, []>("out_17_cast_fp16")]; | |
| tensor<fp16, [512]> input_25_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_25_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74590656)))]; | |
| tensor<fp16, [512]> input_25_beta_0_to_fp16 = const()[name = tensor<string, []>("input_25_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74591744)))]; | |
| tensor<fp16, []> input_25_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_25_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("input_25_cast_fp16")]; | |
| tensor<string, []> input_27_pad_type_0 = const()[name = tensor<string, []>("input_27_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> input_27_strides_0 = const()[name = tensor<string, []>("input_27_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> input_27_pad_0 = const()[name = tensor<string, []>("input_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> input_27_dilations_0 = const()[name = tensor<string, []>("input_27_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> input_27_groups_0 = const()[name = tensor<string, []>("input_27_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [2048, 512, 1, 1]> layers_2_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74592832)))]; | |
| tensor<fp16, [2048]> layers_2_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76690048)))]; | |
| tensor<fp16, [1, 2048, 1, 1]> 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<string, []>("input_27_cast_fp16")]; | |
| tensor<string, []> input_29_mode_0 = const()[name = tensor<string, []>("input_29_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 2048, 1, 1]> input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")]; | |
| tensor<string, []> hidden_states_7_pad_type_0 = const()[name = tensor<string, []>("hidden_states_7_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> hidden_states_7_strides_0 = const()[name = tensor<string, []>("hidden_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = tensor<string, []>("hidden_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> hidden_states_7_dilations_0 = const()[name = tensor<string, []>("hidden_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> hidden_states_7_groups_0 = const()[name = tensor<string, []>("hidden_states_7_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 2048, 1, 1]> layers_2_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76694208)))]; | |
| tensor<fp16, [512]> layers_2_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78791424)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("hidden_states_7_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = tensor<string, []>("inputs_19_cast_fp16")]; | |
| tensor<int32, []> var_714 = const()[name = tensor<string, []>("op_714"), val = tensor<int32, []>(3)]; | |
| tensor<int32, [1]> out_19_axes_0 = const()[name = tensor<string, []>("out_19_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_740_to_fp16 = const()[name = tensor<string, []>("op_740_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_740_to_fp16, x = inputs_19_cast_fp16)[name = tensor<string, []>("out_19_cast_fp16")]; | |
| tensor<fp16, [512]> obj_43_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_43_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78792512)))]; | |
| tensor<fp16, [512]> obj_43_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_43_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78793600)))]; | |
| tensor<fp16, []> obj_43_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_43_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_43_cast_fp16")]; | |
| tensor<string, []> query_13_pad_type_0 = const()[name = tensor<string, []>("query_13_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_13_strides_0 = const()[name = tensor<string, []>("query_13_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_13_pad_0 = const()[name = tensor<string, []>("query_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_13_dilations_0 = const()[name = tensor<string, []>("query_13_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_13_groups_0 = const()[name = tensor<string, []>("query_13_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78794688)))]; | |
| tensor<fp16, [512]> layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79319040)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_13_cast_fp16")]; | |
| tensor<string, []> current_key_7_pad_type_0 = const()[name = tensor<string, []>("current_key_7_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_key_7_strides_0 = const()[name = tensor<string, []>("current_key_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_key_7_pad_0 = const()[name = tensor<string, []>("current_key_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_key_7_dilations_0 = const()[name = tensor<string, []>("current_key_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_key_7_groups_0 = const()[name = tensor<string, []>("current_key_7_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79320128)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_key_7_cast_fp16")]; | |
| tensor<string, []> current_value_7_pad_type_0 = const()[name = tensor<string, []>("current_value_7_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_value_7_strides_0 = const()[name = tensor<string, []>("current_value_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_value_7_pad_0 = const()[name = tensor<string, []>("current_value_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_value_7_dilations_0 = const()[name = tensor<string, []>("current_value_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_value_7_groups_0 = const()[name = tensor<string, []>("current_value_7_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79844480)))]; | |
| tensor<fp16, [512]> layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80368832)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_value_7_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_778_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_778_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_780_cast_fp16 = mul(x = var_51_cast_fp16_3, y = var_137_cast_fp16)[name = tensor<string, []>("op_780_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_13_cast_fp16 = add(x = var_778_cast_fp16, y = var_780_cast_fp16)[name = tensor<string, []>("key_13_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_782_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_782_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_784_cast_fp16 = mul(x = var_60_cast_fp16_3, y = var_137_cast_fp16)[name = tensor<string, []>("op_784_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_13_cast_fp16 = add(x = var_782_cast_fp16, y = var_784_cast_fp16)[name = tensor<string, []>("value_13_cast_fp16")]; | |
| tensor<int32, [4]> var_787 = const()[name = tensor<string, []>("op_787"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_13_cast_fp16 = reshape(shape = var_787, x = query_13_cast_fp16)[name = tensor<string, []>("mh_q_13_cast_fp16")]; | |
| tensor<fp16, []> var_789_to_fp16 = const()[name = tensor<string, []>("op_789_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_790_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_789_to_fp16)[name = tensor<string, []>("op_790_cast_fp16")]; | |
| tensor<int32, [4]> var_791 = const()[name = tensor<string, []>("op_791"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_792_cast_fp16 = reshape(shape = var_791, x = key_13_cast_fp16)[name = tensor<string, []>("op_792_cast_fp16")]; | |
| tensor<bool, []> mh_w_19_transpose_x_0 = const()[name = tensor<string, []>("mh_w_19_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_19_transpose_y_0 = const()[name = tensor<string, []>("mh_w_19_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_790_cast_fp16, y = var_792_cast_fp16)[name = tensor<string, []>("mh_w_19_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_21_cast_fp16 = add(x = mh_w_19_cast_fp16, y = var_155_cast_fp16)[name = tensor<string, []>("mh_w_21_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_800_cast_fp16 = softmax(axis = var_714, x = mh_w_21_cast_fp16)[name = tensor<string, []>("op_800_cast_fp16")]; | |
| tensor<int32, [4]> var_801 = const()[name = tensor<string, []>("op_801"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_802_cast_fp16 = reshape(shape = var_801, x = value_13_cast_fp16)[name = tensor<string, []>("op_802_cast_fp16")]; | |
| tensor<bool, []> attn_13_transpose_x_0 = const()[name = tensor<string, []>("attn_13_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_13_transpose_y_0 = const()[name = tensor<string, []>("attn_13_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_802_cast_fp16, y = var_800_cast_fp16)[name = tensor<string, []>("attn_13_cast_fp16")]; | |
| tensor<int32, [4]> var_805 = const()[name = tensor<string, []>("op_805"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_31_cast_fp16 = reshape(shape = var_805, x = attn_13_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")]; | |
| tensor<string, []> obj_49_pad_type_0 = const()[name = tensor<string, []>("obj_49_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_49_strides_0 = const()[name = tensor<string, []>("obj_49_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_49_pad_0 = const()[name = tensor<string, []>("obj_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_49_dilations_0 = const()[name = tensor<string, []>("obj_49_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_49_groups_0 = const()[name = tensor<string, []>("obj_49_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80369920)))]; | |
| tensor<fp16, [512]> layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80894272)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_49_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_49_cast_fp16)[name = tensor<string, []>("inputs_21_cast_fp16")]; | |
| tensor<int32, [1]> out_21_axes_0 = const()[name = tensor<string, []>("out_21_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_827_to_fp16 = const()[name = tensor<string, []>("op_827_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_827_to_fp16, x = inputs_21_cast_fp16)[name = tensor<string, []>("out_21_cast_fp16")]; | |
| tensor<fp16, [512]> obj_51_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_51_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80895360)))]; | |
| tensor<fp16, [512]> obj_51_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_51_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80896448)))]; | |
| tensor<fp16, []> obj_51_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_51_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_51_cast_fp16")]; | |
| tensor<string, []> query_15_pad_type_0 = const()[name = tensor<string, []>("query_15_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_15_strides_0 = const()[name = tensor<string, []>("query_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_15_pad_0 = const()[name = tensor<string, []>("query_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_15_dilations_0 = const()[name = tensor<string, []>("query_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_15_groups_0 = const()[name = tensor<string, []>("query_15_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80897536)))]; | |
| tensor<fp16, [512]> layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81421888)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_15_cast_fp16")]; | |
| tensor<string, []> key_15_pad_type_0 = const()[name = tensor<string, []>("key_15_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> key_15_strides_0 = const()[name = tensor<string, []>("key_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> key_15_pad_0 = const()[name = tensor<string, []>("key_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> key_15_dilations_0 = const()[name = tensor<string, []>("key_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> key_15_groups_0 = const()[name = tensor<string, []>("key_15_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81422976)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("key_15_cast_fp16")]; | |
| tensor<string, []> value_15_pad_type_0 = const()[name = tensor<string, []>("value_15_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> value_15_strides_0 = const()[name = tensor<string, []>("value_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> value_15_pad_0 = const()[name = tensor<string, []>("value_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> value_15_dilations_0 = const()[name = tensor<string, []>("value_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> value_15_groups_0 = const()[name = tensor<string, []>("value_15_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81947328)))]; | |
| tensor<fp16, [512]> layers_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82471680)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("value_15_cast_fp16")]; | |
| tensor<int32, [4]> var_862 = const()[name = tensor<string, []>("op_862"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_15_cast_fp16 = reshape(shape = var_862, x = query_15_cast_fp16)[name = tensor<string, []>("mh_q_15_cast_fp16")]; | |
| tensor<fp16, []> var_864_to_fp16 = const()[name = tensor<string, []>("op_864_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_865_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_864_to_fp16)[name = tensor<string, []>("op_865_cast_fp16")]; | |
| tensor<int32, [4]> var_866 = const()[name = tensor<string, []>("op_866"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_867_cast_fp16 = reshape(shape = var_866, x = key_15_cast_fp16)[name = tensor<string, []>("op_867_cast_fp16")]; | |
| tensor<bool, []> mh_w_23_transpose_x_0 = const()[name = tensor<string, []>("mh_w_23_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_23_transpose_y_0 = const()[name = tensor<string, []>("mh_w_23_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 1500]> mh_w_23_cast_fp16 = matmul(transpose_x = mh_w_23_transpose_x_0, transpose_y = mh_w_23_transpose_y_0, x = var_865_cast_fp16, y = var_867_cast_fp16)[name = tensor<string, []>("mh_w_23_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_55_cast_fp16 = softmax(axis = var_714, x = mh_w_23_cast_fp16)[name = tensor<string, []>("obj_55_cast_fp16")]; | |
| tensor<int32, [4]> var_871 = const()[name = tensor<string, []>("op_871"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_872_cast_fp16 = reshape(shape = var_871, x = value_15_cast_fp16)[name = tensor<string, []>("op_872_cast_fp16")]; | |
| tensor<bool, []> attn_15_transpose_x_0 = const()[name = tensor<string, []>("attn_15_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_15_transpose_y_0 = const()[name = tensor<string, []>("attn_15_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_872_cast_fp16, y = obj_55_cast_fp16)[name = tensor<string, []>("attn_15_cast_fp16")]; | |
| tensor<int32, [4]> var_875 = const()[name = tensor<string, []>("op_875"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_33_cast_fp16 = reshape(shape = var_875, x = attn_15_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; | |
| tensor<string, []> obj_53_pad_type_0 = const()[name = tensor<string, []>("obj_53_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_53_strides_0 = const()[name = tensor<string, []>("obj_53_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_53_pad_0 = const()[name = tensor<string, []>("obj_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_53_dilations_0 = const()[name = tensor<string, []>("obj_53_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_53_groups_0 = const()[name = tensor<string, []>("obj_53_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82472768)))]; | |
| tensor<fp16, [512]> layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82997120)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_53_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_53_cast_fp16)[name = tensor<string, []>("inputs_23_cast_fp16")]; | |
| tensor<int32, [1]> out_23_axes_0 = const()[name = tensor<string, []>("out_23_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_896_to_fp16 = const()[name = tensor<string, []>("op_896_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_896_to_fp16, x = inputs_23_cast_fp16)[name = tensor<string, []>("out_23_cast_fp16")]; | |
| tensor<fp16, [512]> input_35_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_35_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82998208)))]; | |
| tensor<fp16, [512]> input_35_beta_0_to_fp16 = const()[name = tensor<string, []>("input_35_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82999296)))]; | |
| tensor<fp16, []> input_35_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_35_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("input_35_cast_fp16")]; | |
| tensor<string, []> input_37_pad_type_0 = const()[name = tensor<string, []>("input_37_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> input_37_strides_0 = const()[name = tensor<string, []>("input_37_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> input_37_pad_0 = const()[name = tensor<string, []>("input_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> input_37_dilations_0 = const()[name = tensor<string, []>("input_37_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> input_37_groups_0 = const()[name = tensor<string, []>("input_37_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [2048, 512, 1, 1]> layers_3_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83000384)))]; | |
| tensor<fp16, [2048]> layers_3_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85097600)))]; | |
| tensor<fp16, [1, 2048, 1, 1]> 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<string, []>("input_37_cast_fp16")]; | |
| tensor<string, []> input_39_mode_0 = const()[name = tensor<string, []>("input_39_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 2048, 1, 1]> input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; | |
| tensor<string, []> hidden_states_9_pad_type_0 = const()[name = tensor<string, []>("hidden_states_9_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> hidden_states_9_strides_0 = const()[name = tensor<string, []>("hidden_states_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> hidden_states_9_pad_0 = const()[name = tensor<string, []>("hidden_states_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> hidden_states_9_dilations_0 = const()[name = tensor<string, []>("hidden_states_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> hidden_states_9_groups_0 = const()[name = tensor<string, []>("hidden_states_9_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 2048, 1, 1]> layers_3_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85101760)))]; | |
| tensor<fp16, [512]> layers_3_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87198976)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("hidden_states_9_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = tensor<string, []>("inputs_25_cast_fp16")]; | |
| tensor<int32, []> var_932 = const()[name = tensor<string, []>("op_932"), val = tensor<int32, []>(3)]; | |
| tensor<int32, [1]> out_25_axes_0 = const()[name = tensor<string, []>("out_25_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_958_to_fp16 = const()[name = tensor<string, []>("op_958_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_958_to_fp16, x = inputs_25_cast_fp16)[name = tensor<string, []>("out_25_cast_fp16")]; | |
| tensor<fp16, [512]> obj_57_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_57_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87200064)))]; | |
| tensor<fp16, [512]> obj_57_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_57_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87201152)))]; | |
| tensor<fp16, []> obj_57_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_57_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_57_cast_fp16")]; | |
| tensor<string, []> query_17_pad_type_0 = const()[name = tensor<string, []>("query_17_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_17_strides_0 = const()[name = tensor<string, []>("query_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_17_pad_0 = const()[name = tensor<string, []>("query_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_17_dilations_0 = const()[name = tensor<string, []>("query_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_17_groups_0 = const()[name = tensor<string, []>("query_17_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87202240)))]; | |
| tensor<fp16, [512]> layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87726592)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_17_cast_fp16")]; | |
| tensor<string, []> current_key_9_pad_type_0 = const()[name = tensor<string, []>("current_key_9_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_key_9_strides_0 = const()[name = tensor<string, []>("current_key_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_key_9_pad_0 = const()[name = tensor<string, []>("current_key_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_key_9_dilations_0 = const()[name = tensor<string, []>("current_key_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_key_9_groups_0 = const()[name = tensor<string, []>("current_key_9_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87727680)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_key_9_cast_fp16")]; | |
| tensor<string, []> current_value_9_pad_type_0 = const()[name = tensor<string, []>("current_value_9_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_value_9_strides_0 = const()[name = tensor<string, []>("current_value_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_value_9_pad_0 = const()[name = tensor<string, []>("current_value_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_value_9_dilations_0 = const()[name = tensor<string, []>("current_value_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_value_9_groups_0 = const()[name = tensor<string, []>("current_value_9_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88252032)))]; | |
| tensor<fp16, [512]> layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88776384)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("current_value_9_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_996_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_996_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_998_cast_fp16 = mul(x = var_51_cast_fp16_4, y = var_137_cast_fp16)[name = tensor<string, []>("op_998_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_17_cast_fp16 = add(x = var_996_cast_fp16, y = var_998_cast_fp16)[name = tensor<string, []>("key_17_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1000_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_1000_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1002_cast_fp16 = mul(x = var_60_cast_fp16_4, y = var_137_cast_fp16)[name = tensor<string, []>("op_1002_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_17_cast_fp16 = add(x = var_1000_cast_fp16, y = var_1002_cast_fp16)[name = tensor<string, []>("value_17_cast_fp16")]; | |
| tensor<int32, [4]> var_1005 = const()[name = tensor<string, []>("op_1005"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_17_cast_fp16 = reshape(shape = var_1005, x = query_17_cast_fp16)[name = tensor<string, []>("mh_q_17_cast_fp16")]; | |
| tensor<fp16, []> var_1007_to_fp16 = const()[name = tensor<string, []>("op_1007_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_1008_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1007_to_fp16)[name = tensor<string, []>("op_1008_cast_fp16")]; | |
| tensor<int32, [4]> var_1009 = const()[name = tensor<string, []>("op_1009"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_1010_cast_fp16 = reshape(shape = var_1009, x = key_17_cast_fp16)[name = tensor<string, []>("op_1010_cast_fp16")]; | |
| tensor<bool, []> mh_w_25_transpose_x_0 = const()[name = tensor<string, []>("mh_w_25_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_25_transpose_y_0 = const()[name = tensor<string, []>("mh_w_25_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_1008_cast_fp16, y = var_1010_cast_fp16)[name = tensor<string, []>("mh_w_25_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_155_cast_fp16)[name = tensor<string, []>("mh_w_27_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_1018_cast_fp16 = softmax(axis = var_932, x = mh_w_27_cast_fp16)[name = tensor<string, []>("op_1018_cast_fp16")]; | |
| tensor<int32, [4]> var_1019 = const()[name = tensor<string, []>("op_1019"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_1020_cast_fp16 = reshape(shape = var_1019, x = value_17_cast_fp16)[name = tensor<string, []>("op_1020_cast_fp16")]; | |
| tensor<bool, []> attn_17_transpose_x_0 = const()[name = tensor<string, []>("attn_17_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_17_transpose_y_0 = const()[name = tensor<string, []>("attn_17_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1020_cast_fp16, y = var_1018_cast_fp16)[name = tensor<string, []>("attn_17_cast_fp16")]; | |
| tensor<int32, [4]> var_1023 = const()[name = tensor<string, []>("op_1023"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_41_cast_fp16 = reshape(shape = var_1023, x = attn_17_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")]; | |
| tensor<string, []> obj_63_pad_type_0 = const()[name = tensor<string, []>("obj_63_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_63_strides_0 = const()[name = tensor<string, []>("obj_63_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_63_pad_0 = const()[name = tensor<string, []>("obj_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_63_dilations_0 = const()[name = tensor<string, []>("obj_63_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_63_groups_0 = const()[name = tensor<string, []>("obj_63_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88777472)))]; | |
| tensor<fp16, [512]> layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89301824)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_63_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_63_cast_fp16)[name = tensor<string, []>("inputs_27_cast_fp16")]; | |
| tensor<int32, [1]> out_27_axes_0 = const()[name = tensor<string, []>("out_27_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_1045_to_fp16 = const()[name = tensor<string, []>("op_1045_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1045_to_fp16, x = inputs_27_cast_fp16)[name = tensor<string, []>("out_27_cast_fp16")]; | |
| tensor<fp16, [512]> obj_65_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_65_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89302912)))]; | |
| tensor<fp16, [512]> obj_65_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_65_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89304000)))]; | |
| tensor<fp16, []> obj_65_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_65_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_65_cast_fp16")]; | |
| tensor<string, []> query_19_pad_type_0 = const()[name = tensor<string, []>("query_19_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_19_strides_0 = const()[name = tensor<string, []>("query_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_19_pad_0 = const()[name = tensor<string, []>("query_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_19_dilations_0 = const()[name = tensor<string, []>("query_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_19_groups_0 = const()[name = tensor<string, []>("query_19_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_4_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89305088)))]; | |
| tensor<fp16, [512]> layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89829440)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_19_cast_fp16")]; | |
| tensor<string, []> key_19_pad_type_0 = const()[name = tensor<string, []>("key_19_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> key_19_strides_0 = const()[name = tensor<string, []>("key_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> key_19_pad_0 = const()[name = tensor<string, []>("key_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> key_19_dilations_0 = const()[name = tensor<string, []>("key_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> key_19_groups_0 = const()[name = tensor<string, []>("key_19_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_4_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89830528)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("key_19_cast_fp16")]; | |
| tensor<string, []> value_19_pad_type_0 = const()[name = tensor<string, []>("value_19_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> value_19_strides_0 = const()[name = tensor<string, []>("value_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> value_19_pad_0 = const()[name = tensor<string, []>("value_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> value_19_dilations_0 = const()[name = tensor<string, []>("value_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> value_19_groups_0 = const()[name = tensor<string, []>("value_19_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_4_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90354880)))]; | |
| tensor<fp16, [512]> layers_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90879232)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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<string, []>("value_19_cast_fp16")]; | |
| tensor<int32, [4]> var_1080 = const()[name = tensor<string, []>("op_1080"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_19_cast_fp16 = reshape(shape = var_1080, x = query_19_cast_fp16)[name = tensor<string, []>("mh_q_19_cast_fp16")]; | |
| tensor<fp16, []> var_1082_to_fp16 = const()[name = tensor<string, []>("op_1082_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_1083_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1082_to_fp16)[name = tensor<string, []>("op_1083_cast_fp16")]; | |
| tensor<int32, [4]> var_1084 = const()[name = tensor<string, []>("op_1084"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_1085_cast_fp16 = reshape(shape = var_1084, x = key_19_cast_fp16)[name = tensor<string, []>("op_1085_cast_fp16")]; | |
| tensor<bool, []> mh_w_29_transpose_x_0 = const()[name = tensor<string, []>("mh_w_29_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_29_transpose_y_0 = const()[name = tensor<string, []>("mh_w_29_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 1500]> mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_1083_cast_fp16, y = var_1085_cast_fp16)[name = tensor<string, []>("mh_w_29_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_69_cast_fp16 = softmax(axis = var_932, x = mh_w_29_cast_fp16)[name = tensor<string, []>("obj_69_cast_fp16")]; | |
| tensor<int32, [4]> var_1089 = const()[name = tensor<string, []>("op_1089"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_1090_cast_fp16 = reshape(shape = var_1089, x = value_19_cast_fp16)[name = tensor<string, []>("op_1090_cast_fp16")]; | |
| tensor<bool, []> attn_19_transpose_x_0 = const()[name = tensor<string, []>("attn_19_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_19_transpose_y_0 = const()[name = tensor<string, []>("attn_19_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1090_cast_fp16, y = obj_69_cast_fp16)[name = tensor<string, []>("attn_19_cast_fp16")]; | |
| tensor<int32, [4]> var_1093 = const()[name = tensor<string, []>("op_1093"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_43_cast_fp16 = reshape(shape = var_1093, x = attn_19_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; | |
| tensor<string, []> obj_67_pad_type_0 = const()[name = tensor<string, []>("obj_67_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_67_strides_0 = const()[name = tensor<string, []>("obj_67_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_67_pad_0 = const()[name = tensor<string, []>("obj_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_67_dilations_0 = const()[name = tensor<string, []>("obj_67_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_67_groups_0 = const()[name = tensor<string, []>("obj_67_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_4_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90880320)))]; | |
| tensor<fp16, [512]> layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91404672)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_67_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_67_cast_fp16)[name = tensor<string, []>("inputs_29_cast_fp16")]; | |
| tensor<int32, [1]> out_29_axes_0 = const()[name = tensor<string, []>("out_29_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_1114_to_fp16 = const()[name = tensor<string, []>("op_1114_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1114_to_fp16, x = inputs_29_cast_fp16)[name = tensor<string, []>("out_29_cast_fp16")]; | |
| tensor<fp16, [512]> input_45_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_45_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91405760)))]; | |
| tensor<fp16, [512]> input_45_beta_0_to_fp16 = const()[name = tensor<string, []>("input_45_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91406848)))]; | |
| tensor<fp16, []> input_45_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_45_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("input_45_cast_fp16")]; | |
| tensor<string, []> input_47_pad_type_0 = const()[name = tensor<string, []>("input_47_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> input_47_strides_0 = const()[name = tensor<string, []>("input_47_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> input_47_pad_0 = const()[name = tensor<string, []>("input_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> input_47_dilations_0 = const()[name = tensor<string, []>("input_47_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> input_47_groups_0 = const()[name = tensor<string, []>("input_47_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [2048, 512, 1, 1]> layers_4_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91407936)))]; | |
| tensor<fp16, [2048]> layers_4_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93505152)))]; | |
| tensor<fp16, [1, 2048, 1, 1]> 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<string, []>("input_47_cast_fp16")]; | |
| tensor<string, []> input_49_mode_0 = const()[name = tensor<string, []>("input_49_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 2048, 1, 1]> input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = input_47_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; | |
| tensor<string, []> hidden_states_11_pad_type_0 = const()[name = tensor<string, []>("hidden_states_11_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> hidden_states_11_strides_0 = const()[name = tensor<string, []>("hidden_states_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> hidden_states_11_pad_0 = const()[name = tensor<string, []>("hidden_states_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> hidden_states_11_dilations_0 = const()[name = tensor<string, []>("hidden_states_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> hidden_states_11_groups_0 = const()[name = tensor<string, []>("hidden_states_11_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 2048, 1, 1]> layers_4_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93509312)))]; | |
| tensor<fp16, [512]> layers_4_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95606528)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("hidden_states_11_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor<string, []>("inputs_31_cast_fp16")]; | |
| tensor<int32, []> var_1150 = const()[name = tensor<string, []>("op_1150"), val = tensor<int32, []>(3)]; | |
| tensor<int32, [1]> out_31_axes_0 = const()[name = tensor<string, []>("out_31_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_1176_to_fp16 = const()[name = tensor<string, []>("op_1176_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1176_to_fp16, x = inputs_31_cast_fp16)[name = tensor<string, []>("out_31_cast_fp16")]; | |
| tensor<fp16, [512]> obj_71_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_71_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95607616)))]; | |
| tensor<fp16, [512]> obj_71_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_71_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95608704)))]; | |
| tensor<fp16, []> obj_71_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_71_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_71_cast_fp16")]; | |
| tensor<string, []> query_21_pad_type_0 = const()[name = tensor<string, []>("query_21_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_21_strides_0 = const()[name = tensor<string, []>("query_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_21_pad_0 = const()[name = tensor<string, []>("query_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_21_dilations_0 = const()[name = tensor<string, []>("query_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_21_groups_0 = const()[name = tensor<string, []>("query_21_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95609792)))]; | |
| tensor<fp16, [512]> layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96134144)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("query_21_cast_fp16")]; | |
| tensor<string, []> current_key_pad_type_0 = const()[name = tensor<string, []>("current_key_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_key_strides_0 = const()[name = tensor<string, []>("current_key_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_key_pad_0 = const()[name = tensor<string, []>("current_key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_key_dilations_0 = const()[name = tensor<string, []>("current_key_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_key_groups_0 = const()[name = tensor<string, []>("current_key_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96135232)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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_5_self_attn_k_proj_weight_to_fp16, x = obj_71_cast_fp16)[name = tensor<string, []>("current_key_cast_fp16")]; | |
| tensor<string, []> current_value_pad_type_0 = const()[name = tensor<string, []>("current_value_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> current_value_strides_0 = const()[name = tensor<string, []>("current_value_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> current_value_pad_0 = const()[name = tensor<string, []>("current_value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> current_value_dilations_0 = const()[name = tensor<string, []>("current_value_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> current_value_groups_0 = const()[name = tensor<string, []>("current_value_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96659584)))]; | |
| tensor<fp16, [512]> layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97183936)))]; | |
| tensor<fp16, [1, 512, 1, 1]> current_value_cast_fp16 = conv(bias = layers_5_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_5_self_attn_v_proj_weight_to_fp16, x = obj_71_cast_fp16)[name = tensor<string, []>("current_value_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1214_cast_fp16 = mul(x = current_key_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_1214_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1216_cast_fp16 = mul(x = var_51_cast_fp16_5, y = var_137_cast_fp16)[name = tensor<string, []>("op_1216_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_21_cast_fp16 = add(x = var_1214_cast_fp16, y = var_1216_cast_fp16)[name = tensor<string, []>("key_21_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1218_cast_fp16 = mul(x = current_value_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("op_1218_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1220_cast_fp16 = mul(x = var_60_cast_fp16_5, y = var_137_cast_fp16)[name = tensor<string, []>("op_1220_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_21_cast_fp16 = add(x = var_1218_cast_fp16, y = var_1220_cast_fp16)[name = tensor<string, []>("value_21_cast_fp16")]; | |
| tensor<int32, [4]> var_1223 = const()[name = tensor<string, []>("op_1223"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_21_cast_fp16 = reshape(shape = var_1223, x = query_21_cast_fp16)[name = tensor<string, []>("mh_q_21_cast_fp16")]; | |
| tensor<fp16, []> var_1225_to_fp16 = const()[name = tensor<string, []>("op_1225_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_1226_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1225_to_fp16)[name = tensor<string, []>("op_1226_cast_fp16")]; | |
| tensor<int32, [4]> var_1227 = const()[name = tensor<string, []>("op_1227"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_1228_cast_fp16 = reshape(shape = var_1227, x = key_21_cast_fp16)[name = tensor<string, []>("op_1228_cast_fp16")]; | |
| tensor<bool, []> mh_w_31_transpose_x_0 = const()[name = tensor<string, []>("mh_w_31_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_31_transpose_y_0 = const()[name = tensor<string, []>("mh_w_31_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_31_cast_fp16 = matmul(transpose_x = mh_w_31_transpose_x_0, transpose_y = mh_w_31_transpose_y_0, x = var_1226_cast_fp16, y = var_1228_cast_fp16)[name = tensor<string, []>("mh_w_31_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_33_cast_fp16 = add(x = mh_w_31_cast_fp16, y = var_155_cast_fp16)[name = tensor<string, []>("mh_w_33_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_1236_cast_fp16 = softmax(axis = var_1150, x = mh_w_33_cast_fp16)[name = tensor<string, []>("op_1236_cast_fp16")]; | |
| tensor<int32, [4]> var_1237 = const()[name = tensor<string, []>("op_1237"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_1238_cast_fp16 = reshape(shape = var_1237, x = value_21_cast_fp16)[name = tensor<string, []>("op_1238_cast_fp16")]; | |
| tensor<bool, []> attn_21_transpose_x_0 = const()[name = tensor<string, []>("attn_21_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_21_transpose_y_0 = const()[name = tensor<string, []>("attn_21_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1238_cast_fp16, y = var_1236_cast_fp16)[name = tensor<string, []>("attn_21_cast_fp16")]; | |
| tensor<int32, [4]> var_1241 = const()[name = tensor<string, []>("op_1241"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_51_cast_fp16 = reshape(shape = var_1241, x = attn_21_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; | |
| tensor<string, []> obj_77_pad_type_0 = const()[name = tensor<string, []>("obj_77_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_77_strides_0 = const()[name = tensor<string, []>("obj_77_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_77_pad_0 = const()[name = tensor<string, []>("obj_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_77_dilations_0 = const()[name = tensor<string, []>("obj_77_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_77_groups_0 = const()[name = tensor<string, []>("obj_77_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97185024)))]; | |
| tensor<fp16, [512]> layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97709376)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_77_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_77_cast_fp16)[name = tensor<string, []>("inputs_33_cast_fp16")]; | |
| tensor<int32, [1]> out_33_axes_0 = const()[name = tensor<string, []>("out_33_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_1263_to_fp16 = const()[name = tensor<string, []>("op_1263_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1263_to_fp16, x = inputs_33_cast_fp16)[name = tensor<string, []>("out_33_cast_fp16")]; | |
| tensor<fp16, [512]> obj_79_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_79_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97710464)))]; | |
| tensor<fp16, [512]> obj_79_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_79_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97711552)))]; | |
| tensor<fp16, []> obj_79_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_79_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_79_cast_fp16")]; | |
| tensor<string, []> query_pad_type_0 = const()[name = tensor<string, []>("query_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> query_strides_0 = const()[name = tensor<string, []>("query_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> query_pad_0 = const()[name = tensor<string, []>("query_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> query_dilations_0 = const()[name = tensor<string, []>("query_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> query_groups_0 = const()[name = tensor<string, []>("query_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_5_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97712640)))]; | |
| tensor<fp16, [512]> layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98236992)))]; | |
| tensor<fp16, [1, 512, 1, 1]> query_cast_fp16 = conv(bias = layers_5_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_5_encoder_attn_q_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = tensor<string, []>("query_cast_fp16")]; | |
| tensor<string, []> key_pad_type_0 = const()[name = tensor<string, []>("key_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> key_strides_0 = const()[name = tensor<string, []>("key_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> key_pad_0 = const()[name = tensor<string, []>("key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> key_dilations_0 = const()[name = tensor<string, []>("key_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> key_groups_0 = const()[name = tensor<string, []>("key_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_5_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98238080)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> 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_5_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_cast_fp16")]; | |
| tensor<string, []> value_pad_type_0 = const()[name = tensor<string, []>("value_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> value_strides_0 = const()[name = tensor<string, []>("value_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> value_pad_0 = const()[name = tensor<string, []>("value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> value_dilations_0 = const()[name = tensor<string, []>("value_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> value_groups_0 = const()[name = tensor<string, []>("value_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_5_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98762432)))]; | |
| tensor<fp16, [512]> layers_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99286784)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> value_cast_fp16 = conv(bias = layers_5_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_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_cast_fp16")]; | |
| tensor<int32, [4]> var_1298 = const()[name = tensor<string, []>("op_1298"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_cast_fp16 = reshape(shape = var_1298, x = query_cast_fp16)[name = tensor<string, []>("mh_q_cast_fp16")]; | |
| tensor<fp16, []> var_1300_to_fp16 = const()[name = tensor<string, []>("op_1300_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_1301_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_1300_to_fp16)[name = tensor<string, []>("op_1301_cast_fp16")]; | |
| tensor<int32, [4]> var_1302 = const()[name = tensor<string, []>("op_1302"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_1303_cast_fp16 = reshape(shape = var_1302, x = key_cast_fp16)[name = tensor<string, []>("op_1303_cast_fp16")]; | |
| tensor<bool, []> mh_w_transpose_x_0 = const()[name = tensor<string, []>("mh_w_transpose_x_0"), val = tensor<bool, []>(true)]; | |
| tensor<bool, []> mh_w_transpose_y_0 = const()[name = tensor<string, []>("mh_w_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1, 1500]> mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_1301_cast_fp16, y = var_1303_cast_fp16)[name = tensor<string, []>("mh_w_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_83_cast_fp16 = softmax(axis = var_1150, x = mh_w_cast_fp16)[name = tensor<string, []>("obj_83_cast_fp16")]; | |
| tensor<int32, [4]> var_1307 = const()[name = tensor<string, []>("op_1307"), val = tensor<int32, [4]>([1, 8, 64, -1])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_1308_cast_fp16 = reshape(shape = var_1307, x = value_cast_fp16)[name = tensor<string, []>("op_1308_cast_fp16")]; | |
| tensor<bool, []> attn_transpose_x_0 = const()[name = tensor<string, []>("attn_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_transpose_y_0 = const()[name = tensor<string, []>("attn_transpose_y_0"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [1, 8, 64, 1]> attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_1308_cast_fp16, y = obj_83_cast_fp16)[name = tensor<string, []>("attn_cast_fp16")]; | |
| tensor<int32, [4]> var_1311 = const()[name = tensor<string, []>("op_1311"), val = tensor<int32, [4]>([1, 512, 1, -1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_53_cast_fp16 = reshape(shape = var_1311, x = attn_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; | |
| tensor<string, []> obj_81_pad_type_0 = const()[name = tensor<string, []>("obj_81_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> obj_81_strides_0 = const()[name = tensor<string, []>("obj_81_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> obj_81_pad_0 = const()[name = tensor<string, []>("obj_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> obj_81_dilations_0 = const()[name = tensor<string, []>("obj_81_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> obj_81_groups_0 = const()[name = tensor<string, []>("obj_81_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 512, 1, 1]> layers_5_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99287872)))]; | |
| tensor<fp16, [512]> layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99812224)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("obj_81_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_81_cast_fp16)[name = tensor<string, []>("inputs_35_cast_fp16")]; | |
| tensor<int32, [1]> out_35_axes_0 = const()[name = tensor<string, []>("out_35_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_1332_to_fp16 = const()[name = tensor<string, []>("op_1332_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1332_to_fp16, x = inputs_35_cast_fp16)[name = tensor<string, []>("out_35_cast_fp16")]; | |
| tensor<fp16, [512]> input_55_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_55_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99813312)))]; | |
| tensor<fp16, [512]> input_55_beta_0_to_fp16 = const()[name = tensor<string, []>("input_55_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99814400)))]; | |
| tensor<fp16, []> input_55_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_55_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("input_55_cast_fp16")]; | |
| tensor<string, []> input_57_pad_type_0 = const()[name = tensor<string, []>("input_57_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> input_57_strides_0 = const()[name = tensor<string, []>("input_57_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> input_57_pad_0 = const()[name = tensor<string, []>("input_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> input_57_dilations_0 = const()[name = tensor<string, []>("input_57_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> input_57_groups_0 = const()[name = tensor<string, []>("input_57_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [2048, 512, 1, 1]> layers_5_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99815488)))]; | |
| tensor<fp16, [2048]> layers_5_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101912704)))]; | |
| tensor<fp16, [1, 2048, 1, 1]> 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<string, []>("input_57_cast_fp16")]; | |
| tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 2048, 1, 1]> input_cast_fp16 = gelu(mode = input_mode_0, x = input_57_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; | |
| tensor<string, []> hidden_states_13_pad_type_0 = const()[name = tensor<string, []>("hidden_states_13_pad_type_0"), val = tensor<string, []>("valid")]; | |
| tensor<int32, [2]> hidden_states_13_strides_0 = const()[name = tensor<string, []>("hidden_states_13_strides_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [4]> hidden_states_13_pad_0 = const()[name = tensor<string, []>("hidden_states_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [2]> hidden_states_13_dilations_0 = const()[name = tensor<string, []>("hidden_states_13_dilations_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, []> hidden_states_13_groups_0 = const()[name = tensor<string, []>("hidden_states_13_groups_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [512, 2048, 1, 1]> layers_5_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101916864)))]; | |
| tensor<fp16, [512]> layers_5_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104014080)))]; | |
| tensor<fp16, [1, 512, 1, 1]> 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_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = tensor<string, []>("inputs_cast_fp16")]; | |
| tensor<int32, [1]> out_axes_0 = const()[name = tensor<string, []>("out_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, []> var_1375_to_fp16 = const()[name = tensor<string, []>("op_1375_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_1375_to_fp16, x = inputs_cast_fp16)[name = tensor<string, []>("out_cast_fp16")]; | |
| tensor<fp16, [512]> hidden_states_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104015168)))]; | |
| tensor<fp16, [512]> hidden_states_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104016256)))]; | |
| tensor<fp16, []> hidden_states_epsilon_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> 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<string, []>("hidden_states_cast_fp16")]; | |
| tensor<int32, [1]> var_1386_axes_0 = const()[name = tensor<string, []>("op_1386_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 512, 1]> var_1386_cast_fp16 = squeeze(axes = var_1386_axes_0, x = hidden_states_cast_fp16)[name = tensor<string, []>("op_1386_cast_fp16")]; | |
| tensor<int32, [3]> var_1389_perm_0 = const()[name = tensor<string, []>("op_1389_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<fp16, [51864]> linear_0_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_0_bias_0_to_fp16"), val = tensor<fp16, [51864]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104017344)))]; | |
| tensor<fp16, [1, 1, 512]> var_1389_cast_fp16 = transpose(perm = var_1389_perm_0, x = var_1386_cast_fp16)[name = tensor<string, []>("transpose_0")]; | |
| tensor<fp16, [1, 1, 51864]> logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_1389_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; | |
| tensor<int32, []> var_1393 = const()[name = tensor<string, []>("op_1393"), val = tensor<int32, []>(1)]; | |
| tensor<bool, []> obj_87_interleave_0 = const()[name = tensor<string, []>("obj_87_interleave_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 3072, 1, 1]> key_cache_updates = concat(axis = var_1393, interleave = obj_87_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_cast_fp16))[name = tensor<string, []>("obj_87_cast_fp16")]; | |
| tensor<int32, []> var_1396 = const()[name = tensor<string, []>("op_1396"), val = tensor<int32, []>(1)]; | |
| tensor<bool, []> obj_89_interleave_0 = const()[name = tensor<string, []>("obj_89_interleave_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 3072, 1, 1]> value_cache_updates = concat(axis = var_1396, interleave = obj_89_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_cast_fp16))[name = tensor<string, []>("obj_89_cast_fp16")]; | |
| tensor<int32, [4]> var_1407_begin_0 = const()[name = tensor<string, []>("op_1407_begin_0"), val = tensor<int32, [4]>([0, 3, 0, 0])]; | |
| tensor<int32, [4]> var_1407_end_0 = const()[name = tensor<string, []>("op_1407_end_0"), val = tensor<int32, [4]>([1, 4, 1, 1500])]; | |
| tensor<bool, [4]> var_1407_end_mask_0 = const()[name = tensor<string, []>("op_1407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1407_cast_fp16 = slice_by_index(begin = var_1407_begin_0, end = var_1407_end_0, end_mask = var_1407_end_mask_0, x = obj_55_cast_fp16)[name = tensor<string, []>("op_1407_cast_fp16")]; | |
| tensor<int32, [4]> var_1410_begin_0 = const()[name = tensor<string, []>("op_1410_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1410_end_0 = const()[name = tensor<string, []>("op_1410_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1410_end_mask_0 = const()[name = tensor<string, []>("op_1410_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1410_squeeze_mask_0 = const()[name = tensor<string, []>("op_1410_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1410_cast_fp16 = slice_by_index(begin = var_1410_begin_0, end = var_1410_end_0, end_mask = var_1410_end_mask_0, squeeze_mask = var_1410_squeeze_mask_0, x = var_1407_cast_fp16)[name = tensor<string, []>("op_1410_cast_fp16")]; | |
| tensor<int32, [4]> var_1425_begin_0 = const()[name = tensor<string, []>("op_1425_begin_0"), val = tensor<int32, [4]>([0, 7, 0, 0])]; | |
| tensor<int32, [4]> var_1425_end_0 = const()[name = tensor<string, []>("op_1425_end_0"), val = tensor<int32, [4]>([1, 8, 1, 1500])]; | |
| tensor<bool, [4]> var_1425_end_mask_0 = const()[name = tensor<string, []>("op_1425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1425_cast_fp16 = slice_by_index(begin = var_1425_begin_0, end = var_1425_end_0, end_mask = var_1425_end_mask_0, x = obj_69_cast_fp16)[name = tensor<string, []>("op_1425_cast_fp16")]; | |
| tensor<int32, [4]> var_1428_begin_0 = const()[name = tensor<string, []>("op_1428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1428_end_0 = const()[name = tensor<string, []>("op_1428_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1428_end_mask_0 = const()[name = tensor<string, []>("op_1428_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1428_squeeze_mask_0 = const()[name = tensor<string, []>("op_1428_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1428_cast_fp16 = slice_by_index(begin = var_1428_begin_0, end = var_1428_end_0, end_mask = var_1428_end_mask_0, squeeze_mask = var_1428_squeeze_mask_0, x = var_1425_cast_fp16)[name = tensor<string, []>("op_1428_cast_fp16")]; | |
| tensor<int32, [4]> var_1443_begin_0 = const()[name = tensor<string, []>("op_1443_begin_0"), val = tensor<int32, [4]>([0, 1, 0, 0])]; | |
| tensor<int32, [4]> var_1443_end_0 = const()[name = tensor<string, []>("op_1443_end_0"), val = tensor<int32, [4]>([1, 2, 1, 1500])]; | |
| tensor<bool, [4]> var_1443_end_mask_0 = const()[name = tensor<string, []>("op_1443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1443_cast_fp16 = slice_by_index(begin = var_1443_begin_0, end = var_1443_end_0, end_mask = var_1443_end_mask_0, x = obj_83_cast_fp16)[name = tensor<string, []>("op_1443_cast_fp16")]; | |
| tensor<int32, [4]> var_1446_begin_0 = const()[name = tensor<string, []>("op_1446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1446_end_0 = const()[name = tensor<string, []>("op_1446_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1446_end_mask_0 = const()[name = tensor<string, []>("op_1446_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1446_squeeze_mask_0 = const()[name = tensor<string, []>("op_1446_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1446_cast_fp16 = slice_by_index(begin = var_1446_begin_0, end = var_1446_end_0, end_mask = var_1446_end_mask_0, squeeze_mask = var_1446_squeeze_mask_0, x = var_1443_cast_fp16)[name = tensor<string, []>("op_1446_cast_fp16")]; | |
| tensor<int32, [4]> var_1461_begin_0 = const()[name = tensor<string, []>("op_1461_begin_0"), val = tensor<int32, [4]>([0, 5, 0, 0])]; | |
| tensor<int32, [4]> var_1461_end_0 = const()[name = tensor<string, []>("op_1461_end_0"), val = tensor<int32, [4]>([1, 6, 1, 1500])]; | |
| tensor<bool, [4]> var_1461_end_mask_0 = const()[name = tensor<string, []>("op_1461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1461_cast_fp16 = slice_by_index(begin = var_1461_begin_0, end = var_1461_end_0, end_mask = var_1461_end_mask_0, x = obj_83_cast_fp16)[name = tensor<string, []>("op_1461_cast_fp16")]; | |
| tensor<int32, [4]> var_1464_begin_0 = const()[name = tensor<string, []>("op_1464_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1464_end_0 = const()[name = tensor<string, []>("op_1464_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1464_end_mask_0 = const()[name = tensor<string, []>("op_1464_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1464_squeeze_mask_0 = const()[name = tensor<string, []>("op_1464_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1464_cast_fp16 = slice_by_index(begin = var_1464_begin_0, end = var_1464_end_0, end_mask = var_1464_end_mask_0, squeeze_mask = var_1464_squeeze_mask_0, x = var_1461_cast_fp16)[name = tensor<string, []>("op_1464_cast_fp16")]; | |
| tensor<int32, [4]> var_1479_begin_0 = const()[name = tensor<string, []>("op_1479_begin_0"), val = tensor<int32, [4]>([0, 7, 0, 0])]; | |
| tensor<int32, [4]> var_1479_end_0 = const()[name = tensor<string, []>("op_1479_end_0"), val = tensor<int32, [4]>([1, 8, 1, 1500])]; | |
| tensor<bool, [4]> var_1479_end_mask_0 = const()[name = tensor<string, []>("op_1479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1479_cast_fp16 = slice_by_index(begin = var_1479_begin_0, end = var_1479_end_0, end_mask = var_1479_end_mask_0, x = obj_83_cast_fp16)[name = tensor<string, []>("op_1479_cast_fp16")]; | |
| tensor<int32, [4]> var_1482_begin_0 = const()[name = tensor<string, []>("op_1482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1482_end_0 = const()[name = tensor<string, []>("op_1482_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1482_end_mask_0 = const()[name = tensor<string, []>("op_1482_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1482_squeeze_mask_0 = const()[name = tensor<string, []>("op_1482_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1482_cast_fp16 = slice_by_index(begin = var_1482_begin_0, end = var_1482_end_0, end_mask = var_1482_end_mask_0, squeeze_mask = var_1482_squeeze_mask_0, x = var_1479_cast_fp16)[name = tensor<string, []>("op_1482_cast_fp16")]; | |
| tensor<int32, []> var_1489 = const()[name = tensor<string, []>("op_1489"), val = tensor<int32, []>(1)]; | |
| tensor<bool, []> var_1490_interleave_0 = const()[name = tensor<string, []>("op_1490_interleave_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 5, 1500]> var_1490_cast_fp16 = concat(axis = var_1489, interleave = var_1490_interleave_0, values = (var_1410_cast_fp16, var_1428_cast_fp16, var_1446_cast_fp16, var_1464_cast_fp16, var_1482_cast_fp16))[name = tensor<string, []>("op_1490_cast_fp16")]; | |
| tensor<bool, []> var_1493 = const()[name = tensor<string, []>("op_1493"), val = tensor<bool, []>(false)]; | |
| tensor<int32, [1]> obj_axes_0 = const()[name = tensor<string, []>("obj_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, [1, 1500]> alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_1493, x = var_1490_cast_fp16)[name = tensor<string, []>("obj_cast_fp16")]; | |
| } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); | |
| } |