CI job. Gpt awq 4

[Narsil]

What does this PR do?

Fixes # (issue)

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.

[danieldk]

This block needs a comment what is computed here. The old code was already difficult to read, but I the way I read it is that it is checking whether g_idx incrementing indices by group (so there is a g_idx, but no activation sorting).

But what is this subtraction doing?

[danieldk]

I think it's subtracting the first index, because say we have g_idx with groupsize = 2

[0 0 1 1 2 2 3 3]

If we have two shards, then it gets broken up into

[0 0 1 1] [2 2 3 3]

And then the check fails for the second shard. Subtracting g_idx[0] then changes this into

[0 0 1 1] [0 0 1 1]

Making the check valid. But there should be a comment to make it easier to understand without knowing all the context.

[Narsil]

YEs this is the core idea of the fix. The logic was only working on TP=1, it's now working on TP>1 (to detect that g_idx is redundant and can be safely ignored)

[danieldk]

Yeah makes sense, would be a good to have a short comment here though so that future-us don't have to re-parse what is going on here.