Skip to content

Split the prefix-cache-scorer plugins#323

Merged
github-actions[bot] merged 1 commit intollm-d:mainfrom
vMaroon:precise-prefix-cache
Aug 31, 2025
Merged

Split the prefix-cache-scorer plugins#323
github-actions[bot] merged 1 commit intollm-d:mainfrom
vMaroon:precise-prefix-cache

Conversation

@vMaroon
Copy link
Member

@vMaroon vMaroon commented Aug 29, 2025

Summary

For a more intuitive UX, the llm-d precise-prefix-cache scorer plugin was unified with the IGW prefix-cache scorer. In a later work, a change in the IGW made it so that plugin names cannot be overridden, therefore this unification no longer works.

It is highly possible that given the precise-prefix-cache-aware scheduling well-lit path example, the unification no longer serves its purpose anyway (redundant).

Changes

  • Split the plugins:
    • prefix-cache-scorer points to the IGW's estimate scorer
    • precise-prefix-cache-scorer is what used to be cache_tracking mode
  • Updated documentation
  • Fixed a makefile issue to unblock linting

Relates Issues

FYI @Gregory-Pereira

@vMaroon vMaroon force-pushed the precise-prefix-cache branch 2 times, most recently from 3f37336 to 0ea8159 Compare August 29, 2025 13:51
@vMaroon vMaroon changed the title Split the precise-prefix-cache-scorer plugin Split the prefix-cache-scorer plugins Aug 29, 2025
@vMaroon vMaroon force-pushed the precise-prefix-cache branch 2 times, most recently from d427a3c to 7372bfb Compare August 29, 2025 15:35
@vMaroon
Copy link
Member Author

vMaroon commented Aug 29, 2025

The wasted time and the force-pushes are due to a typo that was fixed in the latest.

- fixed typos install for `make lint`

Signed-off-by: Maroon Ayoub <maroon.ayoub@ibm.com>
@vMaroon vMaroon force-pushed the precise-prefix-cache branch from 7372bfb to 7e4ad00 Compare August 31, 2025 14:27
@shmuelk
Copy link
Collaborator

shmuelk commented Aug 31, 2025

/lgtm

/approve

@github-actions github-actions bot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Aug 31, 2025
@github-actions github-actions bot merged commit aecee68 into llm-d:main Aug 31, 2025
5 checks passed
@vMaroon vMaroon deleted the precise-prefix-cache branch August 31, 2025 15:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

lgtm "Looks good to me", indicates that a PR is ready to be merged.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

make lint don't works in macOS Rename precise prefix scorer to avoid name conflicts with IGW prefix scorer

2 participants