Skip to content

fix: reserve 128 MiB headroom when selecting monolithic execution - #2046

Merged
leejet merged 1 commit into
masterfrom
fix/monolithic-planning-headroom
Sep 24, 2026
Merged

leejet merged 1 commit into
masterfrom
fix/monolithic-planning-headroom

Conversation

@leejet

@leejet leejet commented Sep 24, 2026

Copy link
Copy Markdown
Owner

Summary

  • Require an additional 128 MiB of device and managed-budget headroom before choosing monolithic execution, reducing failures caused by small allocation estimate changes.

Related Issue / Discussion

Fix #2042.

Additional Information

N/A

Checklist

@leejet
leejet merged commit 1a2330d into master Sep 24, 2026
10 checks passed
@leejet
leejet deleted the fix/monolithic-planning-headroom branch September 24, 2026 17:55
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.

CUDA: monolithic workspace check missed by 0.64 MB with all weights resident; no eviction or segmented fallback (Qwen-Image-2.1 edit, 1792x2368, 16 GB)

1 participant