CODEWITHSUNDEEP
nlptransformer-model
박용찬
박용찬

Reputation: 1

Transformer TFBertForSequenceClassification Error(compute_loss() got an unexpected keyword argument 'labels')

2days ago, this code worked. but it don't work now.

I don't know why this code is not working

Please write in details how to work normally this code. (Please)

here is my code and Error. (I used huggingface Transformer API)

Data => Tensor (I used tf.data.Dataset.from_tensor_slices) import tensorflow as tf

with tf.device('/device:GPU:0'):
  train_dataset = tf.data.Dataset.from_tensor_slices((
    dict(train_tokenizer),
    train_target
    ))
  
  val_dataset = tf.data.Dataset.from_tensor_slices((
    dict(val_tokenizer),
    val_target
    ))
  
  test_dataset = tf.data.Dataset.from_tensor_slices((
      dict(test_tokenizer),
      test_target
    ))

Trainer Code

    for epoch in range(5,11):
  training_args = TFTrainingArguments( 
      ...
  )


  def get_model():
    with training_args.strategy.scope():
      config = ...
      model = TFBertForSequenceClassification.from_pretrained(BERT_MODEL, config=config, from_pt=True)
    return model

  model = get_model()

  def compute_metrics(eval_preds):
    metric = load_metric("glue", "mrpc")
    logits, labels = eval_preds
    predictions = np.argmax(logits, axis=-1)
    return metric.compute(predictions=predictions, references=labels)

  trainer = TFTrainer(                    
      model= model,               
      args=training_args,                  
      train_dataset=train_dataset,         
      eval_dataset=val_dataset,            
      compute_metrics = compute_metrics    

  trainer.train()
  trainer.evaluate()

Error Code

in user code:

    File "/usr/local/lib/python3.7/dist-packages/transformers/models/bert/modeling_tf_bert.py", line 1455, in call  *
        loss = None if inputs["labels"] is None else self.compute_loss(labels=inputs["labels"], logits=logits)

    TypeError: compute_loss() got an unexpected keyword argument 'labels'

Call arguments received:
  • input_ids={'input_ids': 'tf.Tensor(shape=(16, 128), dtype=int32)', 'token_type_ids': 'tf.Tensor(shape=(16, 128), dtype=int32)', 'attention_mask': 'tf.Tensor(shape=(16, 128), dtype=int32)'}
  • attention_mask=None
  • token_type_ids=None
  • position_ids=None
  • head_mask=None
  • inputs_embeds=None
  • output_attentions=None
  • output_hidden_states=None
  • return_dict=None
  • labels=tf.Tensor(shape=(16,), dtype=int32)
  • training=True
  • kwargs=<class 'inspect._empty'>

Upvotes: 0

Views: 1037

Answers (1)

MalteHB
MalteHB

Reputation: 11

It seems to be a bug in transformers==4.15.0 at least.

I fixed it by upgrading to transformers==4.16.2.

Upvotes: 1

Related Questions