运行vllm报错：data did not match any variant of untagged enum PyPreTokenizerTypeWrapper

情况描述

环境：

linux
transformers 4.41.2
tokenizers 0.19.1
torch 2.3.0
vllm 0.4.3

在使用vllm运行xverse/XVERSE-13B-256K时(代码如下)：

from vllm import LLM, SamplingParams

llm = LLM(
    model=args.pretrain, 
    trust_remote_code=True, seed=args.seed
)

报错如下

File "/cfs/xxx/xxx.py", line 155, in <module>
    vllm_main(args)
  File "/cfs/xxx/xxx.py", line 84, in vllm_main
    llm = LLM(
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/vllm/entrypoints/llm.py", line 144, in __init__
    self.llm_engine = LLMEngine.from_engine_args(
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/vllm/engine/llm_engine.py", line 359, in from_engine_args
    engine = cls(
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/vllm/engine/llm_engine.py", line 212, in __init__
    self.tokenizer = self._init_tokenizer()
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/vllm/engine/llm_engine.py", line 408, in _init_tokenizer
    return get_tokenizer_group(self.parallel_config.tokenizer_pool_config,
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/vllm/transformers_utils/tokenizer_group/__init__.py", line 20, in get_tokenizer_group
    return TokenizerGroup(**init_kwargs)
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/vllm/transformers_utils/tokenizer_group/tokenizer_group.py", line 23, in __init__
    self.tokenizer = get_tokenizer(self.tokenizer_id, **tokenizer_config)
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/vllm/transformers_utils/tokenizer.py", line 92, in get_tokenizer
    tokenizer = AutoTokenizer.from_pretrained(
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/transformers/models/auto/tokenization_auto.py", line 880, in from_pretrained
    return tokenizer_class.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/transformers/tokenization_utils_base.py", line 2110, in from_pretrained
    return cls._from_pretrained(
  File "/data/miniconda3/envs/xxx/lib/python3.10/site-packages/transformers/tokenization_utils_base.py", line 2336, in _from_pretrained
    tokenizer = cls(*init_inputs, **init_kwargs)
  File "/data/miniconda3/envs/openrlhf/lib/python3.10/site-packages/transformers/tokenization_utils_fast.py", line 114, in __init__
    fast_tokenizer = TokenizerFast.from_file(fast_tokenizer_file)
Exception: data did not match any variant of untagged enum PyPreTokenizerTypeWrapper at line 78 column 3

排查与解决

将xverse/XVERSE-13B-256K换成其他模型后能够正常运行，怀疑是XVERSE-13B-256K和一些版本的库不兼容。由于主要和tokenizer有关，遂降低tokenizer版本：

pip install transformers==4.38.2
pip install tokenizers==0.15.2

成功解决！