Skip to content

fix(ci): use stable cache key for CUDA flash-attn builds#895

Merged
starpit merged 1 commit into
IBM:mainfrom
starpit:fix/cuda-cache-strategy
Feb 20, 2026
Merged

fix(ci): use stable cache key for CUDA flash-attn builds#895
starpit merged 1 commit into
IBM:mainfrom
starpit:fix/cuda-cache-strategy

Conversation

@starpit
Copy link
Copy Markdown
Member

@starpit starpit commented Feb 20, 2026

Summary

  • Remove cargo registry cache step (JS actions run on host in container jobs, cannot access the container ~/.cargo)
  • Use stable cache key based on features + compute cap instead of Cargo.lock hash, so the cache actually hits across releases
  • On exact key match, actions/cache skips the expensive Post save step entirely (~20GB upload avoided)
  • Cargo fingerprinting still recompiles changed deps regardless of cache contents

Test plan

  • Trigger a release build — first run will miss and save (one-time cost)
  • Trigger a second release build — should hit cache, Post save should be skipped
  • Verify flash-attn binaries are correct (deps rebuilt as needed by Cargo fingerprinting)

Generated with Claude Code

The previous cache key included hashFiles Cargo.lock, which changes
every release, causing the cache to always miss and the expensive
Post save step to always run (~20GB upload for no benefit).

- Remove registry cache step (JS actions run on host in container
  jobs, cannot access the container ~/.cargo)
- Use stable key based on features + compute cap so the cache
  actually hits across releases
- On exact match, actions/cache skips Post save entirely
- Cargo fingerprinting still recompiles changed deps regardless

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Nick Mitchell <nickm@us.ibm.com>
@starpit starpit merged commit 944f36c into IBM:main Feb 20, 2026
36 checks passed
@starpit starpit deleted the fix/cuda-cache-strategy branch February 20, 2026 14:14
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant