TrainingArguments are now Immutable

[canberk17]

We have to change the way TrainingArguments are modified due to a recent update that made them immutable, as discussed in this thread.

Previously, we could directly modify the TrainingArguments like this:

training_args.logging_steps = len(corpora_encoded["train"]) // batch_size
training_args.output_dir = "xlm-roberta-base-finetuned-panx-all"

However, since TrainingArguments are now immutable(#25435), we have to use the replace function from the dataclasses module to modify its fields. The replace function creates a new object of the same type as the original, replacing fields with new values specified, as mentioned here

So, the modification of TrainingArguments will now look like this:

training_args = replace(training_args, logging_steps=len(corpora_encoded['train'])//batch_size)
training_args = replace(training_args, output_dir="xlm-roberta-base-finetuned-panx-all")

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[canberk17]

@lvwerra @lewtun

[lvwerra]

The book has the transformers version pinned, so I believe that should not be an issue?

[canberk17]

Aah, fair point. I probably missed that detail while going through the chapter. I'll close this PR. Thanks for pointing it out. Also amazing content!