Transformers: Error arises when using pipeline with community model

[ankandrew]

🐛 Bug

Information

Model I am using is: mrm8488/distill-bert-base-spanish-wwm-cased-finetuned-spa-squad2-es

Language I am using the model on: Spanish

The problem arises when using:

• [ ] the official example scripts:
• [x] my own modified scripts:

The tasks I am working on is:

• [ ] an official GLUE/SQUaD task:
• [x] my own task or dataset:

Steps to reproduce the behavior:

from transformers import *

# Build a pipeline for QA
nlp = pipeline('question-answering', model='mrm8488/distill-bert-base-spanish-wwm-cased-finetuned-spa-squad2-es',
               tokenizer='mrm8488/distill-bert-base-spanish-wwm-cased-finetuned-spa-squad2-es')

nlp(
    {
      'question': 'que queso es?',
      'context': 'Se utilizo en el dia de hoy un queso Emmental'
    }
)

This was working two days ago.

Error log

```html
convert squad examples to features: 0%| | 0/1 [00:00WARNING:transformers.tokenization_utils:Disabled padding because no padding token set (pad_token: [PAD], pad_token_id: 1).
To remove this error, you can add a new pad token and then resize model embedding:
tokenizer.pad_token = ''

model.resize_token_embeddings(len(tokenizer))

RemoteTraceback Traceback (most recent call last)
RemoteTraceback:
"""
Traceback (most recent call last):
File "/usr/lib/python3.6/multiprocessing/pool.py", line 119, in worker
result = (True, func(args, *kwds))
File "/usr/lib/python3.6/multiprocessing/pool.py", line 44, in mapstar
return list(map(args))
File "/usr/local/lib/python3.6/dist-packages/transformers/data/processors/squad.py", line 141, in squad_convert_example_to_features
truncation_strategy="only_second" if tokenizer.padding_side == "right" else "only_first",
File "/usr/local/lib/python3.6/dist-packages/transformers/tokenization_utils.py", line 1796, in encode_plus
*kwargs,
File "/usr/local/lib/python3.6/dist-packages/transformers/tokenization_utils.py", line 1722, in batch_encode_plus
tokens = self._tokenizer.encode(*batch_text_or_text_pairs[0])
File "/usr/local/lib/python3.6/dist-packages/tokenizers/implementations/base_tokenizer.py", line 141, in encode
return self._tokenizer.encode(sequence, pair)
TypeError
"""

The above exception was the direct cause of the following exception:

TypeError Traceback (most recent call last)
in ()
8 nlp({
9 'question': question,
---> 10 'context': context
11 })
12 )

11 frames
/usr/local/lib/python3.6/dist-packages/tokenizers/implementations/base_tokenizer.py in encode()
139 An Encoding
140 """
--> 141 return self._tokenizer.encode(sequence, pair)
142
143 def encode_batch(self, sequences: List[Union[str, Tuple[str, str]]]) -> List[Encoding]:

TypeError:
```

Environment info

• transformers version: 2.5.0
• Python version:3.6.9
• Torch version (GPU?): 1.4.0, running on CPU

[mfuntowicz]

Hi @ankandrew,

Thanks for reporting the issue. Effectively, the QA pipeline is not compatible with fast tokenizers for technical reasons (and I'm currently working on a fix for this).

As a workaround for now, you can disable fast tokenizers when allocating the pipeline:

nlp = pipeline(
    'question-answering', 
    model='mrm8488/distill-bert-base-spanish-wwm-cased-finetuned-spa-squad2-es',
    tokenizer=(
        'mrm8488/distill-bert-base-spanish-wwm-cased-finetuned-spa-squad2-es',  
        {"use_fast": False}
    )
)

nlp(
    {
        'question': 'que queso es?',
        'context': 'Se utilizo en el dia de hoy un queso Emmental'
    }
)
> {'score': 0.36319364208159755, 'start': 31, 'end': 44, 'answer': 'queso Emmental'}

[julien-c]

Also cc'ing @mrm8488 for information while it's in the process of being fixed

[mrm8488]

Thank for the information!

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.