[BUG/Help] <title> 单机多卡部署ptuning微调后模型报错

[ZhaoyingAC]

Is there an existing issue for this?

• [X] I have searched the existing issues

Current Behavior

参考： https://github.com/THUDM/ChatGLM-6B#%E5%A4%9A%E5%8D%A1%E9%83%A8%E7%BD%B2 main_infer.py就是main.py文件中新增加载模型时在参数分布在多gpu上。device_map中新增'transformer.prefix_encoder': 0

    def auto_configure_device_map(num_gpus: int) -> Dict[str, int]:
        # transformer.word_embeddings 占用1层
        # transformer.final_layernorm 、prefix_encoder、lm_head 占用1层
        # transformer.layers 占用 28 层
        # 总共30层分配到num_gpus张卡上
        num_trans_layers = 28
        per_gpu_layers = 30 / num_gpus

        # bugfix: 在linux中调用torch.embedding传入的weight,input不在同一device上,导致RuntimeError
        # windows下 model.device 会被设置成 transformer.word_embeddings.device
        # linux下 model.device 会被设置成 lm_head.device
        # 在调用chat或者stream_chat时,input_ids会被放到model.device上
        # 如果transformer.word_embeddings.device和model.device不同,则会导致RuntimeError
        # 因此这里将transformer.word_embeddings,transformer.final_layernorm,lm_head都放到第一张卡上
        device_map = {'transformer.word_embeddings': 0,
                      'transformer.prefix_encoder': 0, 'transformer.final_layernorm': 0, 'lm_head': 0}

        used = 2
        gpu_target = 0
        for i in range(num_trans_layers):
            if used >= per_gpu_layers:
                gpu_target += 1
                used = 0
            assert gpu_target < num_gpus
            device_map[f'transformer.layers.{i}'] = gpu_target
            used += 1

        return device_map

    if model_args.ptuning_checkpoint is not None:
        # Evaluation
        # Loading extra state dict of prefix encoder
        model = AutoModel.from_pretrained(model_args.model_name_or_path, config=config, trust_remote_code=True)
        prefix_state_dict = torch.load(os.path.join(model_args.ptuning_checkpoint, "pytorch_model.bin"))
        new_prefix_state_dict = {}
        for k, v in prefix_state_dict.items():
            if k.startswith("transformer.prefix_encoder."):
                new_prefix_state_dict[k[len("transformer.prefix_encoder."):]] = v
        model.transformer.prefix_encoder.load_state_dict(new_prefix_state_dict)
    else:
        model = AutoModel.from_pretrained(model_args.model_name_or_path, config=config, trust_remote_code=True)

    if model_args.quantization_bit is not None:
        print(f"Quantized to {model_args.quantization_bit} bit")
        model = model.quantize(model_args.quantization_bit)
    if model_args.pre_seq_len is not None:
        # P-tuning v2
        model = model.half()
        model.transformer.prefix_encoder.float()
    else:
        # Finetune
        model = model.float()
    
    logger.info(model)
    num_gpus = data_args.num_gpus
    if num_gpus >= 2:
        device_map = auto_configure_device_map(num_gpus=num_gpus)
        model = dispatch_model(model, device_map=device_map)

报错信息：
05/05/2023 15:32:06 - INFO - __main__ - ChatGLMForConditionalGeneration(
  (transformer): ChatGLMModel(
    (word_embeddings): Embedding(130528, 4096)
    (layers): ModuleList(
      (0): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (1): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (2): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (3): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (4): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (5): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (6): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (7): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (8): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (9): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (10): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (11): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (12): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (13): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (14): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (15): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (16): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (17): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (18): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (19): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (20): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (21): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (22): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (23): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (24): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (25): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (26): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
      (27): GLMBlock(
        (input_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (attention): SelfAttention(
          (rotary_emb): RotaryEmbedding()
          (query_key_value): Linear(in_features=4096, out_features=12288, bias=True)
          (dense): Linear(in_features=4096, out_features=4096, bias=True)
        )
        (post_attention_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
        (mlp): GLU(
          (dense_h_to_4h): Linear(in_features=4096, out_features=16384, bias=True)
          (dense_4h_to_h): Linear(in_features=16384, out_features=4096, bias=True)
        )
      )
    )
    (final_layernorm): LayerNorm((4096,), eps=1e-05, elementwise_affine=True)
    (prefix_encoder): PrefixEncoder(
      (embedding): Embedding(128, 229376)
    )
    (dropout): Dropout(p=0.1, inplace=False)
  )
  (lm_head): Linear(in_features=4096, out_features=130528, bias=False)
)
05/05/2023 15:32:16 - INFO - __main__ - model.hf_device_map: {'transformer.word_embeddings': 0, 'transformer.prefix_encoder': 0, 'transformer.final_layernorm': 0, 'lm_head': 0, 'transformer.layers.0': 0, 'transformer.layers.1': 0, 'transformer.layers.2': 0, 'transformer.layers.3': 0, 'transformer.layers.4': 0, 'transformer.layers.5': 0, 'transformer.layers.6': 0, 'transformer.layers.7': 0, 'transformer.layers.8': 0, 'transformer.layers.9': 0, 'transformer.layers.10': 0, 'transformer.layers.11': 0, 'transformer.layers.12': 0, 'transformer.layers.13': 1, 'transformer.layers.14': 1, 'transformer.layers.15': 1, 'transformer.layers.16': 1, 'transformer.layers.17': 1, 'transformer.layers.18': 1, 'transformer.layers.19': 1, 'transformer.layers.20': 1, 'transformer.layers.21': 1, 'transformer.layers.22': 1, 'transformer.layers.23': 1, 'transformer.layers.24': 1, 'transformer.layers.25': 1, 'transformer.layers.26': 1, 'transformer.layers.27': 1}
05/05/2023 15:32:16 - INFO - __main__ - model.device: cuda:0

Traceback (most recent call last):
  File "/opt/\*\*\*/ptuning/main_infer.py", line 513, in <module>
    main()
  File "/opt/\*\*\*/ptuning/main_infer.py", line 440, in main
    predict_results = trainer.predict(predict_dataset, metric_key_prefix="predict", max_length=512,
  File "/opt/\*\*\*/ptuning/trainer_seq2seq.py", line 136, in predict
    return super().predict(test_dataset, ignore_keys=ignore_keys, metric_key_prefix=metric_key_prefix)
  File "/opt/\*\*\*/ptuning/trainer.py", line 3020, in predict
    output = eval_loop(
  File "/opt/\*\*\*/ptuning/trainer.py", line 3125, in evaluation_loop
    loss, logits, labels = self.prediction_step(model, inputs, prediction_loss_only, ignore_keys=ignore_keys)
  File "/opt/\*\*\*/ptuning/trainer_seq2seq.py", line 202, in prediction_step
    generated_tokens = self.model.generate(**gen_kwargs)
  File "/root/miniconda3/lib/python3.9/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "/root/miniconda3/lib/python3.9/site-packages/transformers/generation/utils.py", line 1452, in generate
    return self.sample(
  File "/root/miniconda3/lib/python3.9/site-packages/transformers/generation/utils.py", line 2468, in sample
    outputs = self(
  File "/root/miniconda3/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/root/miniconda3/lib/python3.9/site-packages/accelerate/hooks.py", line 165, in new_forward
    output = old_forward(*args, **kwargs)
  File "/root/.cache/huggingface/modules/transformers_modules/chatglm-6b/modeling_chatglm.py", line 1158, in forward
    transformer_outputs = self.transformer(
  File "/root/miniconda3/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/root/.cache/huggingface/modules/transformers_modules/chatglm-6b/modeling_chatglm.py", line 971, in forward
    layer_ret = layer(
  File "/root/miniconda3/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/root/miniconda3/lib/python3.9/site-packages/accelerate/hooks.py", line 165, in new_forward
    output = old_forward(*args, **kwargs)
  File "/root/.cache/huggingface/modules/transformers_modules/chatglm-6b/modeling_chatglm.py", line 609, in forward
    attention_input = self.input_layernorm(hidden_states)
  File "/root/miniconda3/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "/root/miniconda3/lib/python3.9/site-packages/accelerate/hooks.py", line 165, in new_forward
    output = old_forward(*args, **kwargs)
  File "/root/miniconda3/lib/python3.9/site-packages/torch/nn/modules/normalization.py", line 189, in forward
    return F.layer_norm(
  File "/root/miniconda3/lib/python3.9/site-packages/torch/nn/functional.py", line 2503, in layer_norm
    return torch.layer_norm(input, normalized_shape, weight, bias, eps, torch.backends.cudnn.enabled)
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0! (when checking argument for argument weight in method wrapper__native_layer_norm)

Expected Behavior

main.py中predict单机多卡进行预测

Steps To Reproduce

main.py中predict如何单机多卡进行预测

Environment

- OS:
- Python:
- Transformers:
- PyTorch:
- CUDA Support (`python -c "import torch; print(torch.cuda.is_available())"`) :

Anything else?

No response

[sslif]

目前预测推理时，加载模型分两步，先加载原来的模型，再加载微调后的小模型，合起来就是微调后的大模型，单GPU预测的话没有问题，但多GPU的话会报错，解决的思路是，将加载的两个模型合并为一个模型，然后再加载合并后的大模型，多GPU预测就不会有问题。

在做推理预测时，先合并到一个大模型

tokenizer = AutoTokenizer.from_pretrained("model_args.model_name_or_path", trust_remote_code=True)
config = AutoConfig.from_pretrained("model_args.model_name_or_path", trust_remote_code=True)
config.pre_seq_len = model_args.pre_seq_len
model = AutoModel.from_pretrained("model_args.model_name_or_path",config=config, trust_remote_code=True).half().cuda()

prefix_state_dict = torch.load(os.path.join("os.path.join(model_args.ptuning_checkpoint", "pytorch_model.bin"))

new_prefix_state_dict = {}
for k, v in prefix_state_dict.items():
    if k.startswith("transformer.prefix_encoder."):
        new_prefix_state_dict[k[len("transformer.prefix_encoder."):]] = v
model.transformer.prefix_encoder.load_state_dict(new_prefix_state_dict)

model = model.half().cuda()
model.transformer.prefix_encoder.float()

tokenizer.save_pretrained(merge_model_dir)
model.save_pretrained(merge_model_dir, max_shard_size='2GB')

utils.py 文件中只在device_map中新增'transformer.prefix_encoder': 0，其他不变。

import os
from typing import Dict, Tuple, Union, Optional

from torch.nn import Module
from transformers import AutoModel


def auto_configure_device_map(num_gpus: int) -> Dict[str, int]:
    # transformer.word_embeddings 占用1层
    # transformer.final_layernorm 和 lm_head 占用1层
    # transformer.layers 占用 28 层
    # 总共30层分配到num_gpus张卡上
    num_trans_layers = 28
    per_gpu_layers = 30 / num_gpus

    # bugfix: 在linux中调用torch.embedding传入的weight,input不在同一device上,导致RuntimeError
    # windows下 model.device 会被设置成 transformer.word_embeddings.device
    # linux下 model.device 会被设置成 lm_head.device
    # 在调用chat或者stream_chat时,input_ids会被放到model.device上
    # 如果transformer.word_embeddings.device和model.device不同,则会导致RuntimeError
    # 因此这里将transformer.word_embeddings,transformer.final_layernorm,lm_head都放到第一张卡上
    device_map = {'transformer.word_embeddings': 0,
                  'transformer.final_layernorm': 0, 'lm_head': 0,
                  'transformer.prefix_encoder': 0}

    used = 2
    gpu_target = 0
    for i in range(num_trans_layers):
        if used >= per_gpu_layers:
            gpu_target += 1
            used = 0
        assert gpu_target < num_gpus
        device_map[f'transformer.layers.{i}'] = gpu_target
        used += 1

    return device_map


def load_model_on_gpus(checkpoint_path: Union[str, os.PathLike], num_gpus: int = 2,
                       device_map: Optional[Dict[str, int]] = None, **kwargs) -> Module:
    if num_gpus < 2 and device_map is None:
        model = AutoModel.from_pretrained(checkpoint_path, trust_remote_code=True, **kwargs).half().cuda()
    else:
        from accelerate import dispatch_model

        model = AutoModel.from_pretrained(checkpoint_path, trust_remote_code=True, **kwargs).half()

        if device_map is None:
            device_map = auto_configure_device_map(num_gpus)

        model = dispatch_model(model, device_map=device_map)

    return model

最后部署多GPU只需要加载合并后的大模型即可

tokenizer = AutoTokenizer.from_pretrained(merge_model_dir, trust_remote_code=True)
model = load_model_on_gpus(merge_model_dir, num_gpus=2)