Fix beta and exponential sigmas + add tests #9954
Conversation
|
that was fast! :) |
|
@vladmandic could you confirm if this is working for you? @hlky this is extraordinarily fast! Thank you so much! |
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
hlky
force-pushed
the
fix-beta-exponential
branch
from
ca157b7 to
f60e096
Compare
|
Also fixes incorrect ordering in some schedulers. |
|
LGTM! |
|
That was a rigorous test, thanks! Wonder what did you use for the collage? :D |
sdnext built-in xyz grid functionality
|
There was a problem hiding this comment.
thanks @hlky!
looked like there were some issues that you fixed here that did not get caught in our previous tests or @vladmandic 's test, I will try to make & run a more complete slow-test on my end and then merge
| @@ -347,11 +347,33 @@ def set_timesteps(self, num_inference_steps: int, device: Union[str, torch.devic | |||
| ) | |||
| sigmas = np.concatenate([sigmas, [sigma_last]]).astype(np.float32) | |||
| elif self.config.use_exponential_sigmas: | |||
| sigmas = self._convert_to_exponential(in_sigmas=sigmas, num_inference_steps=self.num_inference_steps) | |||
| log_sigmas = np.log(sigmas) | |||
There was a problem hiding this comment.
ohh so it looked like it was wrong before, thank!
* Fix beta and exponential sigmas + add tests --------- Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
What does this PR do?
log_sigmasmoved out ofkarrasbranch_convert_to_exponentialand_convert_to_betareturnnp.ndarrayfor consistency with_convert_to_karrasand other codenum_inference_stepsfor input to_convert_to_exponentialand_convert_to_beta, not all schedulers updateself.num_inference_stepsFixes #9951
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.
@yiyixuxu @vladmandic