-
Notifications
You must be signed in to change notification settings - Fork 50
[acc fix]Change the environment set of DeepSeek-R1 FP4 scripts. #813
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
| vllm serve $model_path \ | ||
| --host localhost \ | ||
| --port 9000 \ | ||
| --port 6789 \ |
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.
can we remain the port number
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.
sure
|
We need rebase the latest origin/dev/perf |
| --seed 123 2>&1 | tee log.server.log & | ||
| --gpu_memory_utilization 0.7 \ | ||
| --block-size 1 \ | ||
| --seed 123 2>&1 | tee log.server.log |
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.
can we add back the &
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.
done
df11d7a to
670b9a4
Compare
|
Is this PR ok to merge? We plan to merge it today. |
Signed-off-by: ZhiweiYan-96 <[email protected]>
This PR fix the Deepseek-r1 MXFP4 accuracy issue. It is based on the fix in PR #809, and uses triton mha for MLA.
Purpose
Test Plan
Test Result
aiter commit: d0a40f55ca1d552f20f2dd55741e7309c936a9d1 (Branch dev/perf)
Eager mode
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.