-
Notifications
You must be signed in to change notification settings - Fork 25.5k
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
[Performance] Keep sigmas on CPU #15823
base: dev
Are you sure you want to change the base?
Conversation
It seems it interferes with #15751? I get
|
That is a problem on their end, and can be resolved by not sending the sigmas to the device. |
* Consistent with implementations in k-diffusion. * Makes this compatible with AUTOMATIC1111#15823
I was testing with hr-fix and it seems it doesn't work?
It only happens when hi-res starts to generate (after tiled upscale process) |
That's not a hi-res fix issue, it's a k-diffusion issue specific to that sampler (Euler and Heun at least, DPM++ 2M works fine). I'll have to work on an upstream patch for k-diffusion for this fix to be viable, then. Marking as draft until then. |
So, for this to resolve, either this PR needs to be merged upstream (crowsonkb/k-diffusion#109) or I have to monkey patch the function. This probably isn't going to be a merge candidate any time soon so I'll give the upstream PR some time. |
Description
Checklist: