search

huggingface / transformers

🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
https://huggingface.co/transformers
Apache License 2.0
134.35k stars 26.86k forks source link

SQUAD Question Answering example:: RuntimeError: Could not infer dtype of NoneType #9843

Closed paniabhisek closed 3 years ago

paniabhisek commented 3 years ago

Environment info

Who can help

NielsRogge commented 3 years ago

I'm not able to reproduce the issue. I went to this page, then clicked on "Open in colab" on the top right (chose PyTorch), and then run the question-answering tutorial, and it's working fine for me:

squad_bis

patil-suraj commented 3 years ago

Hi @paniabhisek

For QA you could use the official run_qa.py example scripts which now supports Trainer and datasets. You can find it here https://github.com/huggingface/transformers/tree/master/examples/question-answering

paniabhisek commented 3 years ago

@NielsRogge I ran the code in colab, it's working for me too. But not in conda environment.

@patil-suraj example-script only supports squad 1.1 ? Does it support squad 2.0 ?

sgugger commented 3 years ago

It supports squad V1 and V2. For V2, just add the flag --version2_with_negative (on top of --dataset_nme squad_v2)

kevinthwu commented 3 years ago

If you try to call train_dataset[137], it returns an error ([136] and [138] both work properly). It is because end_positions.append(encodings.char_to_token(i, answers[i]['answer_end'])) and end_positions.append(encodings.char_to_token(i, answers[i]['answer_end'] + 1)) do not find the correct token; the end_position[-1] is None. The code before #9378 should work.

paniabhisek commented 3 years ago

#9378-comment have worked for me. I was wondering how to use a snippet without an unfamiliar script so I can use my own language model. thanks @kevinthwu .

btw thanks @sgugger I can use the squad 2.0 with the option --version2_with_negative.

I'm not closing as the docs are not updated yet.

BatMrE commented 3 years ago

It supports squad V1 and V2. For V2, just add the flag --version2_with_negative (on top of --dataset_nme squad_v2)

the argument name is 'version_2_with_negative' (line 444 run_qa.py)

github-actions[bot] commented 3 years ago

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.