-
bash tools/check_all.sh(or--cpu-onlywithout a GPU): paste the summary lines - llama.cpp changes:
patches/llama.cpp-extended-strata-moe.patchregenerated (see CONTRIBUTING.md) - a new check was seen to fail on a deliberately broken version
- speed claims: interleaved rounds (
tools/bench_cache.py), with the hardware and thread counts - docs updated (README, docs/reference.md, docs/guide.md) where behavior or flags changed