A cross-platform high-performance inference and training accelerator for ML models with hardware acceleration and graph optimizations.