拥抱面孔训练器:模型初始化错误

Question

我收到以下错误：

You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
Traceback (most recent call last):
  File "./run_hyperparameter_search.py", line 74, in <module>
    trainer = Trainer(
  File "/ext3/miniconda3/lib/python3.8/site-packages/transformers/trainer.py", line 273, in __init__
    model = self.call_model_init()
  File "/ext3/miniconda3/lib/python3.8/site-packages/transformers/trainer.py", line 737, in call_model_init
    raise RuntimeError("model_init should have 0 or 1 argument.")
RuntimeError: model_init should have 0 or 1 argument.
~                                                                       

这是我在我的拥抱面孔训练器中所说的：

#Initialising the model
trainer = Trainer(
    args = training_args,
    tokenizer = tokenizer,
    train_dataset = train_data,
    eval_dataset = val_data,
    # maybe there is a () in the init, but not in compute metrics for sure. Will test
    model_init = finetuning_utils.model_init(),
    compute_metrics = finetuning_utils.compute_metrics,
)

问题显然出在model_init中。

下面是finetuning_utils.model_init()包含的内容：

def model_init():
    """Returns an initialized model for use in a Hugging Face Trainer."""
    ## TODO: Return a pretrained RoBERTa model for sequence classification.
    ## See https://huggingface.co/transformers/model_doc/roberta.html#robertaforsequenceclassification.
    model = RobertaForSequenceClassification.from_pretrained("roberta-base")
    #model = model.to('cuda')
    return model

请帮助解决此错误。

Answer 1

从Huggingface trainer docs看起来，model_init接受了一个可调用的。因此，它不应该实例化参数，而应该接受可调用参数本身，即不带括号：

model_init = finetuning_utils.model_init

或者，您可以删除model_init并使用model参数来达到与finetuning_utils.model_init中包含的代码相同的效果，如下所示：

model = RobertaForSequenceClassification.from_pretrained("roberta-base")