Skip to content

fix: guard GPU memory capacity and propagate encoding failures - #1958

Merged
leejet merged 5 commits into
masterfrom
fix/gpu-memory-failure-handling
Sep 13, 2026
Merged

leejet merged 5 commits into
masterfrom
fix/gpu-memory-failure-handling

Conversation

@leejet

@leejet leejet commented Sep 10, 2026

Copy link
Copy Markdown
Owner

Summary

  • Reserve 512 MiB for backend scratch buffers and pipelines, reject invalid free-memory reports, and bound availability by tracked residency.
  • Return failures from Flux CLIP/T5 encoding and stop image generation before sampling. Log exceptions escaping graph execution with backend context.

Related Issue / Discussion

Try to fix #1946.

Additional Information

N/A

Checklist

@leejet
leejet merged commit 9a97738 into master Sep 13, 2026
10 checks passed
@leejet
leejet deleted the fix/gpu-memory-failure-handling branch September 13, 2026 15:49
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

1 participant