Possible simplification in T5 docs

[xhluca]

I noticed the following nested list comprehension being recommended in the T5 docs:

# replace padding token id's of the labels by -100
labels = [
           [(label if label != tokenizer.pad_token_id else -100) for label in labels_example] for labels_example in labels
] 
labels = torch.tensor(labels)

I believe it's possible to simplify above to:

labels = torch.tensor(labels)
labels[labels == tokenizer.pad_token_id] = -100

I would be happy to make a PR if the change is deemed relevant, though I'm not sure where the docs are located so please let me know if needed :)

[NielsRogge]

Hey,

that's indeed a way more readable way of replacing padding tokens by -100.

The docs of T5 is located here. Note that we are migrating all docs from .rst files to Markdown. But for now, you can add just update the code example.

[xhluca]

Thanks for sharing! I made a PR