Added handling for length <2 of suppress_tokens for whisper

[andreystarenky]

Added handling for <2 length of suppress_tokens for whisper

What does this PR do?

Encountered the following error when implementing a fine-tuned whisper model:

site-packages/transformers/models/whisper/generation_whisper.py", line 1783, in _prepare_decoder_input_ids
    prev_start_of_text = suppress_tokens[-2]
                         ~~~~~~~~~~~~~~~^^^^
IndexError: index -2 is out of bounds for dimension 0 with size 0

The current code assumes that if suppress_tokens is not None it has at least 2 elements. This throws IndexError when suppress_tokens is empty but not None (or length 1) as below:

prev_start_of_text = suppress_tokens[-2] if suppress_tokens is not None else None

I believe the initial code may have intended to check for empty list (if suppress_tokens would have worked) although this would still miss the case of length 1.

My change checks that the list is not None and has a length greater than or equal to 2 (needed to reference second last elem as per original code)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

My apologies if I am tagging the wrong individuals, using list from comments
Models:

• speech models: @ylacombe, @eustlb
  Library:
• generate: @gante

[gante]

LGTM, it would produce an indexing error otherwise 🤗

cc @eustlb I'm merging as this seems like a reasonable change, lmk if you think otherwise :)