Can we add CUDA support to the Rhasspy3 HA Addon?

[nikito]

Hello,
I was attempting to enable CUDA support on the HA addon to test out larger faster-whisper models with CUDA acceleration, but seems the docker image isn't able to support it. I get the following error:
RuntimeError: CUDA failed with error CUDA driver version is insufficient for CUDA runtime version

Thanks! :)

[synesthesiam]

I could build a separate version based on the NVIDIA image. I wasn't aware you could enable GPU access in HA. Is that a separate config.yaml option?

[nikito]

So I think it is specific to the docker running (I think the frigate addon can do it for example, as well as the go2rtc addon used for camera transcode acceleration). I just turned it on in the rhasspy config on the faster-whisper entry when I forked the repository by changing device from "cpu" to "cuda":

My knowledge of docker is a bit limited as I haven't played with that much before (I'm more of a backend developer in c#/Java so a bit new to python and docker lol) so apologies if I am missing anything!

[nikito]

One other thing I realized, I wouldn't be able to try out other models anyway as I don't think they are part of the docker image 😆 Guess the same goes for trying other TTS engines and such?

[kds69]

Hi, did you finally manage to support CUDA?