I am trying to using TVM to tuning models on a arm SOC. The SOC can run the model a spectial hardware, which can accelerate the inference process. However, the hardware needs a special model format which can only be converted from caffe1.0 model. The SOC also provides a cross complier to complie its program. So, I hope to know that is there a way for tvm to export the tuned model into onnx or caffe model?I found that there is a repository called Relay2Onnx but I didn’t find how to use it. And in the source code tvm, the “/python/relay/frontedn”, it seems that there are only methods to import instead of export.
