onnxruntime

Cross-platform, high performance ML inferencing and training accelerator

ONNX Runtime is Microsoft's cross-platform machine learning inference and training accelerator. It provides a C/C++/Python API plus optional execution providers for CPU, NVIDIA CUDA, AMD ROCm, OpenVINO, DirectML, CoreML and others. It reads ONNX-format models and dispatches operators to whichever provider is registered. USE=python builds the Python bindings (numpy + flatbuffers + sympy + coloredlogs); USE=test pulls dev-cpp/gtest for the C++ test suite. Adapted from ::guru at the same version; the ebuild and three patches (relax the flatbuffers dependency, drop -Werror, use system libraries instead of the vendored copies) are imported with upstream authorship headers preserved. ::guru's "prevent generation of PIE" patch is deliberately not carried: it forces POSITION_INDEPENDENT_CODE OFF on every executable target, which weakens ASLR. Eigen3 is still fetched as a specific GitLab archive commit because onnxruntime's CMake requires Eigen fixes that post-date the 3.4.0 tag and predate the 5.x API break. The pre-fetched, hash-pinned distfile is handed to CMake via FETCHCONTENT_SOURCE_DIR_EIGEN3, which also keeps FetchContent from reaching the network mid-build.

homepage ↗homepage ↗ github: microsoft/onnxruntime gitlab: libeigen/eigen