Arthur comments

Results 795 comments of


                                            Arthur

Falcon port

feel free to ping me for a review anytime!

Add training support for EnCodec

Sure! Feel free to open a PR and ping me

Add training support for EnCodec

Seems like he did not link a PR, feel free to synch and ping me for any help! Even a draft is good!

Add training support for EnCodec

The idea is mostly to integrate the loss computation for the VQVAE! Trainer might not work as the model does not use attention, but the target should be to have...

Add training support for EnCodec

Makes sense to me! I think you can focus simply on returning the loss for the modules. The order of training is not that important (when implementing the module wise...

Add training support for EnCodec

Sure, would you mind opening a proper PR? Would be easier to test locally and visualize and follow changes!

`.to_dict` does not correctly serialize `torch.dtype` in some cases (e.g., vision models)

Hey! Indeed, the `PretrainedConfig` class calls `dict_torch_dtype_to_str`, and the `text_config` and `vision_config` inherit from it, so they work fine, indeed, the parent's `torch_dtype` attribute can be modified and we don't...

`.to_dict` does not correctly serialize `torch.dtype` in some cases (e.g., vision models)

Yep, sorry I'll try to get to the original fix taking the comment into account!

`.to_dict` does not correctly serialize `torch.dtype` in some cases (e.g., vision models)

This was fixed by #25237

forced_decoder_ids in Whisper models significantly impacts performance, use decoder_input_ids instead

Hey! Thanks for taking the time to open this PR. Totally get the speedup and the latency induced by the use of `foced_decoder_ids` rather than `decoder_input_ids`. The addition of the...