Repository navigation
Remove the push of the pinned llama.cpp commit to the fork - #569
Conversation
mirror-llama-source pushed every release's pinned upstream commit to bernardladenthin/llama.cpp as jllama/b<nnnn> and needed a write token for the fork (LLAMA_MIRROR_TOKEN) that was never provisioned. The owner maintains that fork, so the job, its script and its tests go. Builds always fetched from upstream and still do; the signed source archive of a release stays and is what rebuilds a release without upstream. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VszvDmRRPP1UcwozsUMY1L
|
Caution Review failedThe pull request is closed. Walkthrough
Merge Risk:
|
|



mirror-llama-sourcepushed the pinned upstream commit of every release tobernardladenthin/llama.cppasjllama/b<nnnn>and neededLLAMA_MIRROR_TOKEN(contents:write on the fork), which was never provisioned -- the first release run would have ended red after publishing. The owner maintains that fork, so the job goes.publish.yml: the job and its mention abovearchive-llama-source;release-gate-exemptions.txt: its line..github/mirror-llama-source.shand.github/buildcheck/tests/test_mirror.pydeleted.llama/CMakeLists.txt:JLLAMA_GIT_REPOSITORYstays (default upstream), the comment no longer names the fork.archive-llama-source,test-source-archive) is unchanged.Builds always fetched from upstream (
ggerganov/llama.cpp->ggml-org/llama.cpp) and still do.Local checks:
releasegate73 jobs, 0 violations; publication wiring 0 problems; run scripts 133, all parse;check-natives.py0 disagreements; workflow parses as YAML; buildcheck tests 97 (7 fewer,test_mirror.py), the same 3 Windows-local failures as onmain.The workspace policy
git-checkout-credentials.mdgets the matching edit in its own PR.🤖 Generated with Claude Code
https://claude.ai/code/session_01VszvDmRRPP1UcwozsUMY1L
Summary by CodeRabbit