Bug #40833: Fix for kv_offset calculation for mixed padding

[preethamyerramsetty]

What does this PR do?

This fixes the kv_offset calculation in cache_utils.py in order to handle left and mixed padding correctly. Previously in case of mixed left and right padding the model could attend to padded tokens which results in incorrect response.

This PR ensures that correct offset is used for left padding, whereas right and mixed paddingfallback to offset 0 and "test_cache_utils.py" is also added to test all padding scenarios.

Fixes #40833

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 - Fixes Generate attends to padded tokens when using left-padded (KV) batches #40833
• 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? - (added unit test in tests/test_cache_utils.py)

Who can review?

Anyone in the community is free to review.

Suggested reviewer: @giulio98

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.

[preethamyerramsetty]

Hi @giulio98, this PR fixes the kv padding issue with Bug #40833 and I have also added the tests for left, right and mixed padding. Can you please review it and get back to me sir. Thank you.

[ArthurZucker]

hello, there is probably an issue because the only diff is string, not code 😅