[WIP] RoPE refactor #39827
Closed
[WIP] RoPE refactor #39827
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Member
Author
|
Keeping the other design version as a cleaner and less breaking option |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Version 1
We want to support defining different RoPE params within one model, similar to how layer types defines different attention patterns. The reason: some models are already using global and local rope params (Gemma3, ModernBert), and we were forced to do monkey patching to support it
This PR is one option how it can be done. Note that it is very much breaking as the RoPE layers will be returning a dict of "cos/sin" when we have several RoPE params. Version 2 is available in #39847, and is less breaking imo
Open for discussion, personally I like version 2