CODEWITHSUNDEEP
pythonpytorchtokenizehuggingface-transformershuggingface-tokenizers
Athena Wisdom
Athena Wisdom

Reputation: 6871

Huggingface error: AttributeError: 'ByteLevelBPETokenizer' object has no attribute 'pad_token_id'

I am trying to tokenize some numerical strings using a WordLevel/BPE tokenizer, create a data collator and eventually use it in a PyTorch DataLoader to train a new model from scratch.

However, I am getting an error

AttributeError: 'ByteLevelBPETokenizer' object has no attribute 'pad_token_id'

when running the following code

from transformers import DataCollatorForLanguageModeling
from tokenizers import ByteLevelBPETokenizer
from tokenizers.pre_tokenizers import Whitespace
from torch.utils.data import DataLoader, TensorDataset

data = ['4814 4832 4761 4523 4999 4860 4699 5024 4788 <unk>']

# Tokenizer
tokenizer = ByteLevelBPETokenizer()
tokenizer.pre_tokenizer = Whitespace()
tokenizer.train_from_iterator(data, vocab_size=1000, min_frequency=1, 
    special_tokens=[
        "<s>",
        "</s>",
        "<unk>",
        "<mask>",
    ])

# Data Collator
data_collator = DataCollatorForLanguageModeling(
    tokenizer=tokenizer, mlm=False
)

train_dataset = TensorDataset(torch.tensor(tokenizer(data, ......)))

# DataLoader
train_dataloader = DataLoader(
    train_dataset, 
    collate_fn=data_collator
)

Is this error due to not having configured the pad_token_id for the tokenizer? If so, how can we do this?

Thanks!

Error trace:

AttributeError: Caught AttributeError in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "/opt/anaconda3/envs/x/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 198, in _worker_loop
    data = fetcher.fetch(index)
  File "/opt/anaconda3/envs/x/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 47, in fetch
    return self.collate_fn(data)
  File "/opt/anaconda3/envs/x/lib/python3.8/site-packages/transformers/data/data_collator.py", line 351, in __call__
    if self.tokenizer.pad_token_id is not None:
AttributeError: 'ByteLevelBPETokenizer' object has no attribute 'pad_token_id'

Conda packages

pytorch                   1.7.0           py3.8_cuda10.2.89_cudnn7.6.5_0    pytorch
pytorch-lightning         1.2.5              pyhd8ed1ab_0    conda-forge
tokenizers                0.10.1                   pypi_0    pypi
transformers              4.4.2                    pypi_0    pypi

Upvotes: 3

Views: 3908

Answers (1)

cronoik
cronoik

Reputation: 19495

The error tells you that the tokenizer needs an attribute called pad_token_id. You can either wrap the ByteLevelBPETokenizer into a class with such an attribute (... and met other missing attributes down the road) or use the wrapper class from the transformers library:

from transformers import PreTrainedTokenizerFast

#your code
tokenizer.save(SOMEWHERE)
tokenizer = PreTrainedTokenizerFast(tokenizer_file=tokenizer_path)

Upvotes: 4

Related Questions