1

我正在尝试使用WordLevel/BPE标记器标记一些数字字符串，创建数据整理器并最终在 PyTorch DataLoader 中使用它来从头开始训练新模型。

但是，我收到一个错误

AttributeError：“ByteLevelBPETokenizer”对象没有属性“pad_token_id”

运行以下代码时

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
)

这个错误是因为没有pad_token_id为分词器配置吗？如果是这样，我们该怎么做？

谢谢！

错误跟踪：

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'

康达包

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
4

1 回答 1

2

该错误告诉您标记器需要一个名为pad_token_id. 您可以将其包装ByteLevelBPETokenizer到具有此类属性的类中（...并在路上遇到其他缺失的属性）或使用转换器库中的包装器类：

from transformers import PreTrainedTokenizerFast

#your code
tokenizer.save(SOMEWHERE)
tokenizer = PreTrainedTokenizerFast(tokenizer_file=tokenizer_path)
于 2021-03-27T16:25:19.713 回答