运行Pegasus改写代码遇'NoneType'不可调用错误，如何规避？

问题重现 #

已安装SentencePiece、sentence-splitter和transformers库，运行以下代码时：

import torch
from transformers import PegasusForConditionalGeneration, PegasusTokenizer
model_name = 'tuner007/pegasus_paraphrase'
torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
tokenizer = PegasusTokenizer.from_pretrained(model_name)
model = PegasusForConditionalGeneration.from_pretrained(model_name).to(torch_device)

def get_response(input_text,num_return_sequences,num_beams):
  batch = tokenizer([input_text],truncation=True,padding='longest',max_length=60, return_tensors="pt").to(torch_device)
  translated = model.generate(**batch,max_length=60,num_beams=num_beams, num_return_sequences=num_return_sequences, temperature=1.5)
  tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
  return tgt_text

出现错误：

TypeError                                 Traceback (most recent call last)
<ipython-input-6-f42eb9e8dd56> in <module>
      5 model_name = 'tuner007/pegasus_paraphrase'
      6 torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
----> 7 tokenizer = PegasusTokenizer.from_pretrained(model_name)
      8 model = PegasusForConditionalGeneration.from_pretrained(model_name).to(torch_device)
      9 

TypeError: 'NoneType' object is not callable

解决方案 #

这个错误源于transformers版本兼容性问题，新版本中PegasusTokenizer的from_pretrained方法已被废弃或指向空对象，可通过以下方式修复：

• 替换为AutoTokenizer：
  修改导入语句和tokenizer初始化代码，用AutoTokenizer自动适配模型对应的tokenizer类型：

  import torch
  from transformers import PegasusForConditionalGeneration, AutoTokenizer
  model_name = 'tuner007/pegasus_paraphrase'
  torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
  tokenizer = AutoTokenizer.from_pretrained(model_name)
  model = PegasusForConditionalGeneration.from_pretrained(model_name).to(torch_device)
  
  def get_response(input_text,num_return_sequences,num_beams):
    batch = tokenizer([input_text],truncation=True,padding='longest',max_length=60, return_tensors="pt").to(torch_device)
    translated = model.generate(**batch,max_length=60,num_beams=num_beams, num_return_sequences=num_return_sequences, temperature=1.5)
    tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
    return tgt_text
  

• 指定兼容的transformers版本：
  如果替换后仍有问题，可安装经过验证的兼容版本：

  pip install transformers==4.28.0
  

• 清理缓存重新下载：
  若模型缓存损坏，删除~/.cache/huggingface/hub目录下对应模型的缓存文件，重新运行代码让系统自动下载完整的模型和tokenizer文件。

内容的提问来源于stack exchange，提问作者Hanan Raihana