[GPU] Fix kernels synchronization in PagedAttention operation #28645
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
sshlyapn
changed the title
…otation is enabled but skipped for the current iteration
sshlyapn
force-pushed
the
fix_paged_attention_sync
branch
from
a920974 to
d4e5cb9
Compare
|
LGTM |
p-durandin
approved these changes
Jan 23, 2025
github-merge-queue bot
pushed a commit
that referenced
this pull request
Jan 23, 2025
### Details: - Fix synchronization in PagedAttention operation when KV-cache rotation is enabled but skipped for the current iteration. Previously, `dep_events` was always replaced with `res_events` if `has_rotated_blocks=true`, leading to empty events vector being passed to the next kernels and causing accuracy deviations in cases of out_of_order queue due to missing synchronization - PR to master: #28645
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Outdated suggestions cannot be applied.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.
Details:
Previously,
dep_eventswas always replaced withres_eventsifhas_rotated_blocks=true, leading to empty events vector being passed to the next kernels and causing accuracy deviations in cases of out_of_order queue due to missing synchronization