Skip to content

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

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
starpit merged 1 commit into from
Feb 20, 2026
Merged

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

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 10 additions & 17 deletions .github/workflows/release-cli.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -448,29 +448,22 @@ jobs:
# Skip flash-attn variants on PRs (only run on release)
if: github.event_name == 'release' || matrix.platform.features == 'cuda'

- name: Cache Cargo registry
# Only use cache for flash-attn variants on releases (they take longer to build)
if: github.event_name == 'release' && matrix.platform.features != 'cuda'
uses: actions/cache@v5
with:
path: |
~/.cargo/registry/index/
~/.cargo/registry/cache/
~/.cargo/git/db/
key: linux-cargo-registry-${{ hashFiles('**/Cargo.lock') }}
restore-keys: |
linux-cargo-registry-

- name: Cache Cargo build
# Only use cache for flash-attn variants on releases (they take longer to build)
# Cache target/ for flash-attn variants on releases (they take ~45min to build).
# Note: ~/.cargo/registry is NOT cached because this is a container job — JS actions
# (like actions/cache) run on the host, which can't access the container's ~/.cargo.
# The target/ dir works because it's in the workspace, which is host-mounted.
#
# Key is stable (no Cargo.lock hash) so the cache actually hits across releases.
# On exact match, actions/cache skips the expensive Post save step entirely.
# Cargo's fingerprinting handles staleness — it recompiles changed deps regardless.
if: github.event_name == 'release' && matrix.platform.features != 'cuda'
uses: actions/cache@v5
with:
path: target/
key: linux-${{ matrix.platform.arch }}-cargo-build-cuda-${{ matrix.platform.features }}-${{ hashFiles('**/Cargo.lock') }}
key: cuda-build-${{ matrix.platform.features }}-${{ matrix.platform.cuda_compute_cap }}
restore-keys: |
linux-${{ matrix.platform.arch }}-cargo-build-cuda-${{ matrix.platform.features }}-
linux-${{ matrix.platform.arch }}-cargo-build-cuda-
cuda-build-${{ matrix.platform.features }}-

- name: Install build dependencies
# Skip flash-attn variants on PRs (only run on release)
Expand Down
Loading