ROPE computing suitable for NPU

Open leisuzz opened this issue 4 weeks ago • 1 comments

What does this PR do?

In this way, the computing time cost for NPU will reduce about 3%-5%.

[ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
[x] Did you read the contributor guideline?
[x] Did you read our philosophy doc (important for complex PRs)?
[ ] 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.
[x] Did you write any new necessary tests?

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.

Dec 13 '25 08:12 leisuzz

@sayakpaul Please take a look at this PR, thank you!

Dec 13 '25 09:12 leisuzz

I will pause this PR for this moment, because it needs to some tests with FSDP2 in FLUX.2. Sorry for the inconvenience.

Dec 18 '25 02:12 leisuzz

@sayakpaul @yiyixuxu The Flux.2 has been tested as well. The performance also increased. Thank you for your time :)

Dec 19 '25 07:12 leisuzz