🔥[Community Event] Doc Tests Sprint - Configuration files🔥

[ydshieh]

This sprint is similar to #16292 - but for model configuration files, i.e. configuration_[model_name].py.
For example, src/transformers/models/bert/configuration_bert.py

The expected changes

The changes we expect could be find #19485:

1. Change the import order of the model and configuration classes
2. Add (with random weights) in the comment before model initialization line
3. Add configuration_[model_name].py to utils/documentation_tests.txt (respecting the order)

Please do step 3. only after Running the doctest and make sure all tests pass (see below) 🙏

How to run doctests

Suppose you are working on src/transformers/models/bert/configuration_bert.py. The steps to run the test are:

0. Stage your changes

   git add src/transformers/models/bert/configuration_bert.py

1. Prepare the files to be tested

   python utils/prepare_for_doc_test.py src

   or if you prefer to be more specific

   python utils/prepare_for_doc_test.py src/transformers/models/bert/configuration_bert.py

   This will change some files (doc-testing needs to add additional lines that we don't include in the doc source files).

2. Launch the test:

   python -m pytest --doctest-modules src/transformers/models/bert/configuration_bert.py -sv --doctest-continue-on-failure

3. Cleanup git status

   git checkout -- .

   to clean up the changes in step 1.

Ready (or not)?

If all tests pass, you can commit, push and open a PR 🔥 🚀 , otherwise iterate the above steps 💯 !

[daspartho]

I'd like to work on this; I'll start with YOLOS and open a PR :)

[daspartho]

I'll take on Whisper!

[daspartho]

I'll work on Beit!

[imarekkus]

FYI Bart #19524 and Albert #19541 are already done :)

[daspartho]

I'll take on GPT2 next!

[RamitPahwa]

I can take imageGPT

[grgkaran03]

i'll work on yoso

[daspartho]

I'll work on

• RoBERTa
• ViT
• DeiT
• Reformer

and open up PR soon :)

[RamitPahwa]

Also will raise for Transformer-XL !

[imarekkus]

I'll work on bloom

[sha016]

I'll take fsmt next

[Saad135]

While browsing the list of model configurations, I noticed that the DebertaConfig class does not have an example docstring section. Unsure if that is supposed to be like that, but just incase its not, I will add a PR to include the example docstring and maybe I can get some feedback from there.

[kushal-gopal]

I'll work on dpt

[ydshieh]

DebertaConfig

That would be very nice, @Saad135 ! Thank you

[Saad135]

I will take DeBERTa-v2 next

[Saad135]

I can take camembert next

[Saad135]

I can take DPR next

[Saad135]

I can take DeformableDetrConfig next

[JuheonChu]

Can I take timesformer next?

[ydshieh]

Can I take timesformer next?

Sure! For the context, we decide not to use the tiny random model checkpoints anymore. If there are some downstream models which lack the checkpoint, we just not to provide the expected values.

[elabongaatuo]

Hello, I would like to take on gptj, longformer, and hubert

[elabongaatuo]

@ydshieh , may I share a list of models that are yet to be worked on?

[ydshieh]

@elabongaatuo GPT-J is large, and our CI won't be able to run doctest with its checkpoints.

I think gptj, longformer, and hubert are all covered in

https://github.com/huggingface/transformers/blob/5f3ea66bc0c27ad2a8761fdf8489cf7d72257b93/utils/documentation_tests.txt

Feel free to check the modeling files that are not in the above file 🤗 if you want to work on it ❤️ . Thank you!

[elabongaatuo]

@ydshieh , thank you. m2m_100,llama and mvp don't have modeling files. a go ahead to work on them?

[ydshieh]

llama has no publicly available checkpoints on the Hub - no need to work on it.
For the other 2 files, you can run doctest against them. If they pass, you can simply add them to documentation_tests.txt.
Otherwise, we can discuss how to deal with the errors :-).

• we might need to use a community user's checkpoint
• or a checkpoint without real weights, and indicate this

[PhalitJotwani]

Hi @ydshieh , I am new to open source, so just wanted to confirm whether I can take Falcon or not? Config file of Falcon is not mentioned in the documentation_tests.txt file.

[AVAniketh0905]

Hello @ydshieh , I am new to open sorce and want to take barthez. If my contributions are successful, I'm eager to extend my involvement to other models as well. Looking forward to a productive and enduring journey of contributions!

Edit: I couldn't find configuration files for barthez. Any help is appreciated!

[hegdeadithyak]

I'll take roformer #26530