Skip to content

Fix recent GEGLU changes slowing down inference unet models#1079

Closed
stduhpf wants to merge 1 commit into
leejet:masterfrom
stduhpf:unet-perf
Closed

Fix recent GEGLU changes slowing down inference unet models#1079
stduhpf wants to merge 1 commit into
leejet:masterfrom
stduhpf:unet-perf

Conversation

@stduhpf

@stduhpf stduhpf commented Dec 12, 2025

Copy link
Copy Markdown
Contributor

Fixes #984

@stduhpf

stduhpf commented Dec 12, 2025

Copy link
Copy Markdown
Contributor Author

@leejet It looks like ggml_ext_chunk() is a serious bottleneck. Probably because of the repeated use of ggml_permute() and ggml_cont(). Maybe rewriting it to avoid relying on permutations instead of merging this PR would be even better. Let me know if you'd prefer me giving it a shot.

@leejet

leejet commented Dec 12, 2025

Copy link
Copy Markdown
Owner

Perhaps #1080 is a simpler fix.

@stduhpf stduhpf closed this Dec 12, 2025
@wbruna wbruna mentioned this pull request Dec 12, 2025
@stduhpf stduhpf deleted the unet-perf branch June 14, 2026 21:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug] Performance regression

2 participants