| # Contributing to TurboQuant Elyan |
|
|
| Thanks for your interest in contributing! |
|
|
| ## How to Contribute |
|
|
| ### Bug Reports |
| 1. Search existing issues to avoid duplicates |
| 2. Open a new issue with clear reproduction steps |
| 3. Include compiler, OS/arch, and whether the VSX or scalar path was used |
|
|
| ### Code Contributions |
| 1. **Fork** this repository |
| 2. **Create a branch**: `git checkout -b feature/your-feature` |
| 3. **Write code** in the style of the file you are touching |
| 4. **Add tests** for new functionality |
| 5. **Submit a PR** with a clear description |
|
|
| ### Code Style |
| - C11 for `ggml-turboquant-vsx.h`, CUDA C++ for `turboquant_cuda_kernel.cu` |
| - 4-space indent, `_tq3_` prefix on internal helpers |
| - The header must keep building with no `-mvsx`: the scalar fallback is the |
| reference implementation |
|
|
| ### Testing |
| ```bash |
| make test # suite + tq3_self_test(), scalar or native VSX |
| make test-vsx # cross-build for POWER8 |
| sh tests/check_backends_agree.sh # header, CUDA kernel and README agree |
| ``` |
| - New features require new tests |
| - All tests must pass on CI |
| - Anything that changes `norm`, `scale` or the packed indices also changes |
| the golden block in `tests/test_turboquant_vsx.c`; regenerate it and say |
| so in the PR, since it is what pins cross-backend bit-exactness |
|
|
| ### Quality Checklist |
| - [ ] Code follows project style |
| - [ ] New code has tests |
| - [ ] All existing tests pass |
| - [ ] No hardcoded secrets or API keys |
| - [ ] Documentation updated if needed |
|
|
| ## Code of Conduct |
| Be respectful. Constructive feedback only. No harassment. |
|
|
| ## License |
| By contributing, you agree your code is licensed under the same license as this project. |
|
|
