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

New torch-directml release needed #545

Open

ananosleep opened this issue Jan 12, 2024 · 1 comment

ananosleep commented Jan 12, 2024

As a Stable Diffusion user, new torch-directml is needed for issue listed here:

bug issue:

bad memory allocation: GPU memory leak after every generation; https://github.com/vladmandic/automatic/discussions/2250#discussioncomment-7113512
produce black or grey tile on bottom left in some ways: https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111/issues/93; not only in this extension, but ScuNET and SwinIR upscale also suffer from this bug.

feature needed:

fp8 storage support, which needed torch 2.1.0+, and could save lots of GPU memory;
upcoming fp8 files loaded support in future, which will able in upcoming torch2.2.

Looking forward.

hez2010 commented May 22, 2024

Seems that a new version of pytorch-directml has been released, with up to pytorch 2.2.1 support.
https://pypi.org/project/torch-directml/0.2.1.dev240521/

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment