grammars: fix resampling logic regression
#7424
Merged
Conversation
This file contains 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
This was referenced May 21, 2024
mscheong01
approved these changes
May 21, 2024
There was a problem hiding this comment.
Yes, I believe this fix is valid. My PR modifies the optimization behavior to apply grammar checks first and then resample without them if necessary(which wouldn't happen at all since grammar checks are already applied) while it should be the opposite. Thank you for addressing this 🙇♂️ 🙇♂️
teleprint-me
pushed a commit
to teleprint-me/llama.cpp
that referenced
this pull request
May 23, 2024
|
Sorry I missed reviewing this before -- this would have been good to get vetted / merged, but now I suspect that this is going to conflict a fair bit with #8643 -- I'm not sure how imminent that PR is, but this one still may be worth pushing through. |
|
I made a note to apply it within #8643 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
I think #6240 may have undone the performance gains of #4306 (just because of extra negation in
!is_resampling)This brings them back / makes grammar sampling fast again! (#4218)
I've tested this along with #7370 (early exit) & #6811 (cache codepoints) & other changes (see benchmarks): looks like only this PR & to a lesser degree, the codepoint caching, are needed / useful.
cc/ @kalomaze, @mscheong01, @HanClinto, @AlienKevin
I'm not sure if the performance benchmark bot would now catch this kind of regression, ideas welcome!