Running Home Assistant wyoming voice assistant with AMD ROCm onnxruntime possible?

Hi,
I am evaluating wether ROCm is a viable alternative to Cuda to buy a AMD RX 9070 XT GPU.
I successfully managed to install rocm on tumbleweed and I can use it with pytorch.

Now I’d like to run wyoming-piper for Home Assistant with AMD GPU support.
I managed to run wyoming-piper (and wyoming-faster-whisper) on tumbleweed, but failed to run it with AMD GPU support.

It seems wyoming-piper uses piper which uses onnxruntime which can use ROCm.
For that onnxruntime MIGraphXExecutionProvider (and/or ROCMExecutionProvider?) are required.
I did not find tumbleweed packages for them and also not for MIGraphX which would be required to build the onnx runtime execution providers from source.

Any hints?
Anyone using piper or whisper with ROCm?

1 Like

I tried for a week but failed. Now bought an Nvidia board :frowning:

I found an AMD provided docker image with a working onnxruntime and ROCmExecutionProvider, but it still won’t work with that.
Error was CUDA versions of driver and runtime would not match.
I tried several version combinations without success.
Too much fiddling for my taste…

1 Like

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.