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

Request to add 4-bit AdamW #208

Open
LiutongZhou opened this issue Sep 13, 2023 · 3 comments
Open

Request to add 4-bit AdamW #208

LiutongZhou opened this issue Sep 13, 2023 · 3 comments
Assignees
Labels
feature request Request features

Comments

@LiutongZhou
Copy link

LiutongZhou commented Sep 13, 2023

Paper and Code

Paper: Memory Efficient Optimizers with 4-bit States
Code : https://github.com/thu-ml/low-bit-optimizers/blob/main/lpmm/optim/optimizer.py

@LiutongZhou LiutongZhou added the feature request Request features label Sep 13, 2023
@kozistr
Copy link
Owner

kozistr commented Sep 17, 2023

thanks for the request! it requires cuda files to compile like bitsandbytes optimizer, so it's complicated to move entire implementations into this project, instead, maybe we can support the optimizer like this example, that manually install or compile, and use it through pytorch-optimizer.

I'll figure it out how to smoothly support this in pytorch-optimizer :)

thank you!

@LiutongZhou
Copy link
Author

LiutongZhou commented Sep 17, 2023

Liked the idea of supporting it like the example.

Maybe even easier, make the FR to bnb and include it here?

Made the FR to TimDettmers/bitsandbytes#775

@kozistr
Copy link
Owner

kozistr commented Sep 18, 2023

Liked the idea of supporting it like the example.

Maybe even easier, make the FR to bnb and include it here?

Made the FR to TimDettmers/bitsandbytes#775

that's a great idea! it'd be much better that bnb supports 4-bit optimizers. then, for now, wait for the response and decide whether including both or only bnb.

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

No branches or pull requests

2 participants