-
-
Notifications
You must be signed in to change notification settings - Fork 7.7k
[Bugfix] Fix async log stats #8417
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
/ready |
👋 Hi! Thank you for contributing to the vLLM project. Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can do one of these:
🚀 |
627c81f
to
048fffc
Compare
Does it fix #8178? |
@DarkLight1337 not sure if it is related to #8178. |
@robertgshaw2-neuralmagic @comaniac |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Otherwise LGTM
Head branch was pushed to by a user without write access
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. I would call it preempted_indices
but it seems like this could be used for other uses in the future
Some of this has been resolved, but when applying the --num-scheduler-steps option, the values are not realistic for some metrics. |
Signed-off-by: Alvant <alvasian@yandex.ru>
Co-authored-by: Alexander Matveev <59768536+alexm-neuralmagic@users.noreply.github.com>
Signed-off-by: LeiWang1999 <leiwang1999@outlook.com>
This PR fixes the issue described here: #8219 (comment)
The fix is simple, we need to skip log stats for sequences that are already preempted since their state is reset to PREFILL (for recompute).