Huggingface转换器模型返回字符串而不是logits

Question

我试着从huggingface网站上运行这个例子。https://huggingface.co/transformers/task_summary.html。该模型似乎返回两个字符串，而不是logits！这会导致torch.argmax()抛出一个错误

    from transformers import AutoTokenizer, AutoModelForQuestionAnswering
    import torch
    
    tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
    
    model = AutoModelForQuestionAnswering.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad", return_dict=True)
    
    text = r"""? Transformers (formerly known as pytorch-transformers and pytorch-pretrained-bert) provides general-purpose
    architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet…) for Natural Language Understanding (NLU) and Natural
    Language Generation (NLG) with over 32+ pretrained models in 100+ languages and deep interoperability between
    TensorFlow 2.0 and PyTorch.
    """
    
    questions = ["How many pretrained models are available in ? Transformers?",
    "What does ? Transformers provide?",
    "? Transformers provides interoperability between which frameworks?"]
    
    for question in questions:
      inputs = tokenizer(question, text, add_special_tokens=True, return_tensors="pt")
      input_ids = inputs["input_ids"].tolist()[0] # the list of all indices of words in question + context
    
      text_tokens = tokenizer.convert_ids_to_tokens(input_ids) # Get the tokens for the question + context
      answer_start_scores, answer_end_scores = model(**inputs)
    
      answer_start = torch.argmax(answer_start_scores)  # Get the most likely beginning of answer with the argmax of the score
      answer_end = torch.argmax(answer_end_scores) + 1  # Get the most likely end of answer with the argmax of the score
    
      answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(input_ids[answer_start:answer_end]))
    
      print(f"Question: {question}")
      print(f"Answer: {answer}")

Answer 1

由于最近的一次更新，模型现在返回特定于任务的输出对象(字典)，而不是普通的元组。您使用的网站尚未更新以反映此更改。您可以通过指定return_dict=False来强制模型返回元组

answer_start_scores, answer_end_scores = model(**inputs, return_dict=False)

或者，您可以通过调用values()方法从QuestionAnsweringModelOutput对象中提取值：

answer_start_scores, answer_end_scores = model(**inputs).values()

或者甚至使用QuestionAnsweringModelOutput对象：

outputs = model(**inputs)
answer_start_scores = outputs.start_logits
answer_end_scores = outputs.end_logits