Skip to content
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.

Sign up for GitHub

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

Jump to bottom

fix bug in deep fp8 zero #141

Merged
merged 4 commits into from
Dec 14, 2023
Merged

fix bug in deep fp8 zero #141

merged 4 commits into from
Dec 14, 2023

Conversation

tocean
Copy link
Contributor

@tocean tocean commented Dec 8, 2023 •
edited
Loading

Description
Fix 2 bugs in fp8 zero optimizer.
MS-AMP will crash when:

  • there are param groups which don't have high precision parameter
  • partition 0 has no parameters.

@tocean tocean requested a review from wkcn December 8, 2023 05:40
wkcn
wkcn reviewed Dec 13, 2023
View reviewed changes
Copy link
Contributor

@wkcn wkcn left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The cases when ref_value is None and ref_scale is None should be handled.

msamp/deepspeed/runtime/zero/fp8_stage_1_and_2.py Outdated Show resolved Hide resolved
msamp/deepspeed/runtime/zero/fp8_stage_1_and_2.py Outdated Show resolved Hide resolved
wkcn
wkcn approved these changes Dec 14, 2023
View reviewed changes
Copy link
Contributor

@wkcn wkcn left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. Thanks!

@tocean tocean merged commit bf6f01a into main Dec 14, 2023
9 checks passed
@tocean tocean deleted the yuxiang/zero_bugfix branch December 14, 2023 05:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@guoshzhao guoshzhao guoshzhao approved these changes

@wkcn wkcn wkcn approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@tocean @wkcn @guoshzhao