Unified Attention Accuracy Bugfixes #393
Merged
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
Signed-off-by: Konrad Zawora <[email protected]>
Signed-off-by: Konrad Zawora <[email protected]>
kzawora-intel
requested review from
adobrzyn,
afierka-intel,
mgawarkiewicz-intel,
michalkuligowski,
mswiniarsk,
vivekgoe and
xuechendi
as code owners
Signed-off-by: Konrad Zawora <[email protected]>
🚧 CI BlockedThe main CI workflow was not started for the following reason:
|
mswiniarsk
approved these changes
Oct 14, 2025
✅ CI PassedAll checks passed successfully against the following vllm commit: |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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Outdated suggestions cannot be applied.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've noticed two accuracy issues in unified attention:
unified_execute_modelmethod.The first one had major impact - I suspect we were malforming batches as the generation process went on, since the self.input_batch.num_tokens & req_state.output_token_ids were not updated correctly - in Granite GSM8K fixing that yielded +10 percentage points improvement
The second one had a negligible impact - I didn't notice any acc improvement in any tests I've run - but we should be masking anything above context length regardless.
I've added GSM8k accuracy test to CI with this PR that should now pass as well.