-
|
I would like to execute some BERT related benchmarks ex: However I confirm I have required execution providers are installed.
anyone knows what's going on ? thanks. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 2 replies
-
|
You need to build ONNX Runtime with ROCM and MIGraphX enabled. See https://github.com/microsoft/onnxruntime/blob/main/dockerfiles/Dockerfile.migraphx#L56 for the build instructions |
Beta Was this translation helpful? Give feedback.
Installing Torch without ROCm can also cause this. i.e. if you ran
pip install torch. That version of torch is fully mapped to cuda. You have two options uninstall torch and install via a rocm repopip install torch==2.0.1 --trusted-host https://repo.radeon.com/ -f https://repo.radeon.com/rocm/manylinux/rocm-rel-5.7/or use the rocm pytorch docker and install ORT in it...rocm/pytorch:latest