I have a model with custom ops (and dynamic axes). What is the best way for me to import this model? I was looking through the relax and relay frontends for ONNX and they both seem like they don’t support custom ops. In my case the custom operators are rendering operators related to rendering textured meshes. I’ve implemented custom TVM operators for these cuda kernels already, I’m just not sure how to associate the onnx operators in the onnx graph to my tvm operators during import?
