mirror of https://github.com/Helldez/BigMoeOnEdge.git synced 2026-10-19 15:33:46 +00:00

chore: retire the files and the words the architecture table left behind (#277 )

#266 replaced the table of architectures with the layout read off the
file, and its changelog entry says recipe.h, arch_registry.cpp and
docs/adding-a-model.md are gone; the commit left them in the tree,
compiled by nothing and linked from nowhere. Remove them, and the four
places that still spoke of recipe rows and --list-archs: the release
note, CONTRIBUTING.md, a comment in session.cpp, and the inspect script,
whose verdict now says what the engine will read off the converted file
instead of printing a row to paste.

2026-10-18 11:33:12 +02:00

2.9 KiB

Raw Permalink Blame History

Contributing

Thanks for your interest. BigMoeOnEdge stays deliberately small and disciplined.

Ground rules

Never patch third_party/llama.cpp in-tree. Work through its public API. Keeping the submodule stock is the point of the project (see docs/seam.md).
The byte-identity gates must pass. Streaming must stay lossless. Run ctest --output-on-failure before opening a PR; if you change the streamer or the seam, say so in the PR and note the gate result.
No hardcoding. There is no table of architectures: which tensors are the routed experts is read from the file (docs/which-models-stream.md). A new expert layout gets a byte-identity gate.
English everywhere. Conventional Commits. Match .clang-format.

Getting set up

git clone --recursive https://github.com/Helldez/BigMoeOnEdge.git
cd BigMoeOnEdge
scripts/build-host.sh
cd build && ctest --output-on-failure

CI also runs the gates under AddressSanitizer and UndefinedBehaviorSanitizer. To reproduce a report locally, build into a separate directory and run the same tests:

SAN="-fsanitize=address,undefined -fno-omit-frame-pointer"
cmake -S . -B build-asan -G Ninja -DCMAKE_BUILD_TYPE=RelWithDebInfo \
  -DCMAKE_C_FLAGS="$SAN" -DCMAKE_CXX_FLAGS="$SAN" \
  -DCMAKE_EXE_LINKER_FLAGS="-fsanitize=address,undefined" \
  -DCMAKE_SHARED_LINKER_FLAGS="-fsanitize=address,undefined"
cmake --build build-asan
cd build-asan && ASAN_OPTIONS=halt_on_error=1 \
  UBSAN_OPTIONS=halt_on_error=1:print_stacktrace=1:suppressions=$PWD/../tests/sanitizers/ubsan.supp \
  ctest --output-on-failure

tests/sanitizers/ubsan.supp and lsan.supp list only findings inside third_party/llama.cpp, which this project does not patch; a report in our own code is a bug to fix, never a line to add there. Leak checking is Linux-only (macOS's ASan has none); on Linux add LSAN_OPTIONS=suppressions=$PWD/../tests/sanitizers/lsan.supp.

The web UI: cd ui && npm ci && npm test && npm run check && npm run build.

Good first contributions

A benchmark on hardware we do not have. No code needed: scripts/bench-report.sh MODEL.gguf runs the fixed protocol and prints the tables; open a benchmark report. See docs/community-benchmarks.md for the hardware list and what a row needs.
A byte-identity gate for an expert layout the gates do not cover yet (docs/which-models-stream.md).
Read coalescing / expert-contiguous layout experiments (see docs/roadmap.md).

Reporting problems

Bugs: the bug report form. Security problems: privately, as SECURITY.md says, never in a public issue.

PRs

Keep them focused. Describe what you measured if the change is performance-related, and on what hardware. Update the relevant doc in the same PR as the code.

2.9 KiB Raw Permalink Blame History