StableHLO

StableHLO is an open-source operation set for machine learning models, acting as a portability layer between ML frameworks (such as TensorFlow, JAX, and PyTorch) and ML compilers (such as XLA and IREE). It is based on the MHLO dialect and adds serialization and versioning using MLIR bytecode, providing backward-compatibility guarantees. The repository includes a specification and C++ and Python implementations for defining StableHLO programs.