AI Tech Suite

ONNC

Click to visit website

About

ONNC (Open Neural Network Compiler) is a specialized compilation framework designed for deep learning accelerators that support ONNX (Open Neural Network Exchange) operators. It enables seamless model transformation into DLA-specific binaries, ensuring that ONNX models execute efficiently across various deep learning accelerators. ONNC, an open-source tool, is notable for its support of NVDLA-based hardware, allowing the compilation of models into NVDLA Loadable files and facilitating system-level exploration of inference designs. It integrates with the LLVM compiler, offering a modular framework that can adapt to both compatible and unique DLA designs. The compiler is structured into distinct phases, providing reusable optimizations and a flexible pass manager, making it accessible for AI researchers and engineers familiar with LLVM. ONNC also supports a vibrant community of developers contributing through various sub-projects, tutorials, and GitHub collaboration.