Hermetic Toolchains for ML

rules_ml_toolchain provides toolchains for ML projects to achieve reproducible, consistent, and isolated builds across Linux, macOS, and Windows.

Hermetic builds benefits:

• Reproducibility: Every build produces identical results regardless of the developer's machine environment.
• Consistency: Eliminates "works on my machine" issues, ensuring builds are consistent across different development environments.
• Isolation: Compilers, linkers, runtimes, interpreters, and sysroots are isolated from the host system, minimizing unexpected dependencies and side effects.

Provided Hermetic Toolchains

This project provides hermetic toolchains for C++, Python, Fortran, and GPU:

• C++ Toolchains (cc/README.md): Hermetic LLVM/Clang C++ toolchains and sysroots for Linux (x86_64, aarch64), macOS (aarch64), and Windows (x86_64), including support for libstdc++ and libc++ linking, sanitizers (ASan, TSan), cross-compilation, and Clang header modules.
• Python Toolchains (py/README.md): Hermetic CPython interpreters (3.11, 3.12, 3.13, 3.14, including free-threaded builds), locked PyPI dependency management, and Pywrap rules (rules_pywrap) for building C++/Python (pybind11) bindings.
• Fortran Toolchains (fortran/README.md): Hermetic LLVM flang toolchains for building Fortran targets and mixed Fortran/C/C++ libraries (such as BLAS and LAPACK) on Linux (x86_64, aarch64), macOS (aarch64), and Windows (x86_64).
• GPU Toolchains (gpu/README.md): Hermetic GPU toolchains and redistributables for NVIDIA CUDA, CUDNN, NCCL, and NVSHMEM, AMD ROCm, and Intel oneAPI SYCL.