Hello, I used a yolov5s model in onnx format with autoscheduler,and the optimized model works fine on the agx-Xaviar, but an error is reported when the same command is used through the xshell(Segmentation fault (core dumped)).
After I used auto to tune my onnx model, the tuned model worked properly on agx, but after logging in to agx via ssh using xshell, I got an error( Segmentation fault (core dumped))when calling the same model file.
How does this kind of problem come about? What is the difference between ssh running and native running? How do I solve this problem?
My python version is 3.7.13, onnx version is 1.11.0, tvm version is 0.9, llvm version is 10.0, cuda version is 10.2, cudNN version is 8.0.0, jetpack version is 4.5