Fix black images when using SageAttention as optimized_attention #784
Conversation
This file contains hidden or 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
pamparamm
force-pushed
the
sageattention_fix
branch
from
956f1ca to
f14af9a
Compare
|
@cubiq Sorry for the inconvenience, but can you please review this PR? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
IPAdapter with precise weight types produces black images during generation when used together with SageAttention - probably due to numerical instability in its quantization algorithms when dealing with near-zero tensors.
This PR adds
epsilonvariable inipadapter_attentionfunction that fixes this issue. I've set it to 0 for alloptimized_attentionexcept forattention_sage.