-
Notifications
You must be signed in to change notification settings - Fork 4.2k
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
Support the parallel conversion from ZeRO checkpoints to FP32/FP16/BF16 param weight #6655
base: master
Are you sure you want to change the base?
Support the parallel conversion from ZeRO checkpoints to FP32/FP16/BF16 param weight #6655
Conversation
…ter weights in parallel via UCP
Hi @xylian86 thanks for the contribution, could you run the pre-commit formatter locally to fix the Formatting error? |
@loadams Thank you for the reminder. I run the pre-commit formatter locally before the PR but It seems that formatting outputs can vary across different environments. After switching to a new machine with ubuntu-22.04 and re-running pre-commit formatter, I've resolved the formatting issues. |
Hi @xylian86 - looks like there are a few merge conflicts now - could you resolve those, then we can get this merged? |
@xylian86 - ping on this if you have time to resolve the merge conflicts we can get this reviewed as well. |
This PR adds parallel conversion support for the zero_to_fp32 script using universal checkpoints, addressing the request at ISSUE#6526. "The previous version was written 3 years ago models were small and converted fast. Now with 70B+ models the conversion can take hours."