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.

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

Add 4-bit quantization and QLoRA #487

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Conversation

jeff52415
Copy link

@jeff52415 jeff52415 commented May 30, 2023

#486

  • Add 4-bit Quantization in Training/Inference Script.
  • Add QLoRA Optimizer.
  • Dependency Update: I've updated our dependencies to their latest stable versions to ensure optimal functionality and security. Please see the updated requirements.txt file for the specific changes.
  • Readme Update: The Readme file has been updated to reflect these changes, and to provide clearer instructions for users and contributors.

Please review the changes at your earliest convenience. Any feedback or suggestions are welcome.

Best Regards,
Jeff Yang

@tloen
Copy link
Owner

tloen commented May 31, 2023

Noting down here: while I'm not actively maintaining this repo, this is one of the PRs I'd like to get tested and merged in the next day or two.

@jeff52415
Copy link
Author

@tloen , if you require any assistance or support, please do not hesitate to inform me.

@shaonianyr
Copy link

@jeff52415 i wonder 4-bit train speed is the same as 8-bit or slower?

@jeff52415
Copy link
Author

jeff52415 commented Jun 7, 2023

@jeff52415 i wonder 4-bit train speed is the same as 8-bit or slower?

While I haven't conducted formal benchmarking to confirm this, my empirical evidence suggests a higher speed for 4-bit training.

@chand1012
Copy link

Is this going to get merged? I found it quite helpful!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants