Sylvain Gugger

The issue is probably in the modeling code missing some `.contiguous()` calls.

Hi there! The datasets attribute of the `Trainer` are never modified, so they always retain all their columns. You therefore don't need to change the training arguments (which is something...

I am very confused as to how letting the callback change the training arguments would help in this instance. By the time you arrive at the model, the dataloader has...

Most likely the model itself, from what you shared. Normally data collators collate what they get (as long as it's in a "collatable" type). As you also pointed out, you...

Thanks for your PR! It would be great to have the `ViTMSNForImageClassification` even if there are no released weights for image classification, so users can already fine-tune the main checkpoint...

You are trying to run the examples of the main branch along with an older version of Transformers. Either upgrade your Transformers or use the [examples matching v4.18.0](https://github.com/huggingface/transformers/tree/v4.18.0/examples).

Which could all go in a `parse_yaml_file` method :-) Doing this and also refactoring the `parse_json_file` to use `parse_dict`, as well as adding small tests would be nice additions that...

No, you can create it during the test by saving some dictionary (look at the `parse_dict` tests) into a temporary file.

You can also use the context manager for a temp dir. ``` with tempfile.TemporaryDirectory() as tmp_dir: # Save file in tmp_dir as usual # do the tests ``` The plus...

I am not sure I fully understand the problem here. It looks like the canonical repos have weights that mismatch our code. If this is the case, those weights should...