EulerAncestral add rescale_betas_zero_snr
#6187
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Uses same infinite sigma fix from EulerDiscrete. Interestingly the ancestral version had the opposite problem: too much contrast instead of too little.
It helps this scheduler too, particularly when the model is using bf16. While the noise dtype is still the model's it's automatically upcasted for the add so all it affects is determinism.
|
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. |
|
@yiyixuxu is this good to be merged? |
|
@sayakpaul LGTM! |
|
Thank you for your contributions! |
yoonseokjin
pushed a commit
to yoonseokjin/diffusers
that referenced
this pull request
Dec 25, 2023
* EulerAncestral add `rescale_betas_zero_snr` Uses same infinite sigma fix from EulerDiscrete. Interestingly the ancestral version had the opposite problem: too much contrast instead of too little. * UT for EulerAncestral `rescale_betas_zero_snr` * EulerAncestral upcast samples during step() It helps this scheduler too, particularly when the model is using bf16. While the noise dtype is still the model's it's automatically upcasted for the add so all it affects is determinism. --------- Co-authored-by: Sayak Paul <[email protected]>
donhardman
pushed a commit
to donhardman/diffusers
that referenced
this pull request
Dec 29, 2023
* EulerAncestral add `rescale_betas_zero_snr` Uses same infinite sigma fix from EulerDiscrete. Interestingly the ancestral version had the opposite problem: too much contrast instead of too little. * UT for EulerAncestral `rescale_betas_zero_snr` * EulerAncestral upcast samples during step() It helps this scheduler too, particularly when the model is using bf16. While the noise dtype is still the model's it's automatically upcasted for the add so all it affects is determinism. --------- Co-authored-by: Sayak Paul <[email protected]>
AmericanPresidentJimmyCarter
pushed a commit
to AmericanPresidentJimmyCarter/diffusers
that referenced
this pull request
Apr 26, 2024
* EulerAncestral add `rescale_betas_zero_snr` Uses same infinite sigma fix from EulerDiscrete. Interestingly the ancestral version had the opposite problem: too much contrast instead of too little. * UT for EulerAncestral `rescale_betas_zero_snr` * EulerAncestral upcast samples during step() It helps this scheduler too, particularly when the model is using bf16. While the noise dtype is still the model's it's automatically upcasted for the add so all it affects is determinism. --------- Co-authored-by: Sayak Paul <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Implements the
rescale_betas_zero_snrfield on EulerAncestralDiscreteScheduler, using the same implementation as regular euler seen in #6024.Once again I upcast the samples for the duration of
step()as it significantly improves the output quality at no performance cost. I leave the ancestral noise as the model's dtype to preserve existing seeds, but this could be changed to float32 as well to make seeds mostly deterministic across model dtypes.#Copy frompoints to DDIM again, and the UT is the same as established ddim/euler.