chore(deps): pin nvidia/cuda base to 12.6.x (#449)

Dependabot proposed nvidia/cuda 12.6.3 to 12.9.2 (#441), a minor bump the existing semver-major ignore did not catch. The 12.6 base is deliberate: the AI bundle wheels are cu126, libcublas-12-6 matches, and a 12.9 base requires driver R575+ so the container fails to start on common 570.x production drivers (the regression PR #334 fixed). Ignore nvidia/cuda >= 12.7 so 12.6.x patch bumps still flow while 12.7+ is held back.
This commit is contained in:
SnapOtter
2026-07-06 20:18:03 +08:00
committed by GitHub
parent d6ae79bebb
commit b28c9dd2da
+8
View File
@@ -44,6 +44,14 @@ updates:
ignore:
- dependency-name: "*"
update-types: ["version-update:semver-major"]
# nvidia/cuda base is pinned to 12.6.x on purpose. The AI bundle wheels are
# cu126 (torch/paddle/onnxruntime-gpu), libcublas-12-6 matches it, and a 12.9
# base requires driver R575+ so the container fails to start on common 570.x
# production drivers. See the comment above the FROM in docker/Dockerfile.
# Allow 12.6.x patch bumps; block 12.7+ until the wheels and driver baseline
# move together (closed dependabot #441 for this reason).
- dependency-name: "nvidia/cuda"
versions: [">= 12.7"]
# GitHub Actions
- package-ecosystem: "github-actions"