Correctly inject config in PytorchModelHubMixin

[Wauplin]

Solves #2079.

With this PR, users don't have to rely on a config parameter to configure their model. We read default+passed values automatically to build the config file ourselves. This way, we are able to save a correct config.json file so that reloaded model have exact same config. Only "jsonable" fields are saved in config.json.

And ofc it's backward compatible so users that prefer to pass config.json directly are still supported.

PR is inspired by Inspired by facebookresearch/hiera#26. cc @dbolya with this change, you won't need the @has_config decorator anymore 🤗

Here is an example to showcase it:

>>> import torch
>>> import torch.nn as nn
>>> from huggingface_hub import PyTorchModelHubMixin

>>> class MyModel(nn.Module, PyTorchModelHubMixin):
...     def __init__(self, hidden_size: int = 512, vocab_size: int = 30000, output_size: int = 4):
...         super().__init__()
...         self.param = nn.Parameter(torch.rand(hidden_size, vocab_size))
...         self.linear = nn.Linear(output_size, vocab_size)

...     def forward(self, x):
...         return self.linear(x + self.param)
>>> model = MyModel(hidden_size=256)

# Save model weights to local directory
>>> model.save_pretrained("my-awesome-model")

# Push model weights to the Hub
>>> model.push_to_hub("my-awesome-model")

# Download and initialize weights from the Hub
>>> model = MyModel.from_pretrained("username/my-awesome-model")
>>> model.hidden_size
256

cc @mfuntowicz as well. This shouldn't impact you except that the generated config.json files should be more complete.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[NielsRogge]

Great! This is going to be very useful

[Wauplin]

That looks great. However, in your PR description, do you mean self.hidden_size instead of config.hidden_size etc. in order to access the attributes?

Yes sorry, I've updated the example.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 89.18919% with 4 lines in your changes are missing coverage. Please review.

Project coverage is 82.91%. Comparing base (05d715f) to head (0c900cd).
Report is 1 commits behind head on main.

Files	Patch %	Lines
src/huggingface_hub/hub_mixin.py	89.18%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2079      +/-   ##
==========================================
+ Coverage   82.90%   82.91%   +0.01%     
==========================================
  Files         102      102              
  Lines        9480     9499      +19     
==========================================
+ Hits         7859     7876      +17     
- Misses       1621     1623       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[LysandreJik]

Looks great! That is sure to come in handy with model serialization, no need to handle a config on the side anymore. Awesome!

[LysandreJik]

yess 🎉

[Wauplin]

Thanks for the review @LysandreJik 🤗 Yes I also expect it to make integrations even more straightforward now :)