ankane / ml-builds

Pre-built shared libraries for machine learning

ML Builds

Pre-built shared libraries for machine learning

Library	Linux x86	ARM	Mac x86	ARM	Windows	Files	License
Bling Fire	✓	✓	✓	✓	✓	View	MIT
cmfrec	✓		✓	✓		View	MIT
Faiss	✓		✓		n/a	View	MIT
Interpret	✓		✓	✓	✓	View	MIT
LIBMF	✓	✓	✓	✓	✓	View	BSD-3-Clause
MITIE	✓		✓	✓	✓	View	BSL-1.0
Multicore t-SNE	✓		✓	✓	✓	View	BSD-3-Clause
NGT	✓	✓	✓	✓	n/a	View	Apache-2.0
OSQP	✓	✓	✓	✓	✓	View	Apache-2.0
Prophet	✓	✓	✓	✓		View	MIT
SCS	✓		✓		✓	View	MIT
ThunderSVM	✓		✓	✓	✓	View	Apache-2.0
XGBoost	✓	✓	✓	✓	✓	View	Apache-2.0
xLearn	✓	n/a	✓	n/a	✓	View	Apache-2.0
Argon2 (non-ML)	✓	✓	✓	✓	✓	View	CC0-1.0

Notes

• LIBMF: Fork, no OpenMP
• Multicore t-SNE: Fork with OpenMP for Mac x86
• SCS: No blas/lapack on Windows
• XGBoost: No OpenMP on Windows

View the build scripts

Some projects include shared libraries as part of their releases:

• LibTorch
• LightGBM
• ONNX Runtime
• OR-Tools

Library	Linux ARM	Mac ARM	Files	License
LightGBM		✓	View	MIT

Details

For portability:

• Linux x86 libraries are built with -march=x86-64 (when possible) and older glibc
• Linux ARM libraries are built with -march=armv8-a (when possible) and older glibc
• Mac x86 libraries are built with -march=nehalem (same as Homebrew) and MACOSX_DEPLOYMENT_TARGET=10.13
• Mac ARM libraries are built with -march=armv8-a and MACOSX_DEPLOYMENT_TARGET=11.0

References:

• x86
• ARM

Other notes:

• Windows libraries are built with -A x64
• Mac libraries that use OpenMP require brew install libomp (Faiss, Multicore t-SNE, NGT, ThunderSVM, XGBoost)