ONNX is a great runtime for universal model files. I've seen TVM benchmarks showing better performance. It would be cool to see if these really do perform better and how hard it would be to integrate into our system.
It looks like it uses DLPack which may mean it would simplify the code a lot!
ONNX is a great runtime for universal model files. I've seen TVM benchmarks showing better performance. It would be cool to see if these really do perform better and how hard it would be to integrate into our system.