Error response from daemon: unknown or invalid runtime name: nvidia

[Yukun-Cui]

After ran docker compose --profile model up, I met follow error.

I have set the /etc/docker/daemon.json.
nvidia-smi could show my GPUs successfully.
My environment : win11 22621.675 + docker desktop + WSL2.0 Ubuntu-20.04

[nicozhan]

Hi @1572161937 !
After successfully installing nvidia runtime, please try to restart Docker.

[Yukun-Cui]

I have restarted win11, restarted docker, restarted WSL2.0, it doesn't work.
But I am not sure whether I have installed nvidia runtime, how can i check it?

[nicozhan]

If the runtime in docker info contains nvidia, the installation is successful. (Refer to nvidia official website for installation.)

[Yukun-Cui]

The issue seems to be:
The Linux server uses the nvidia-docker to start CUDA's Docker. While in WSL2.0, the docker run --gpus is used.
However, in the docker-hub package, this module is packaged as a whole and cannot modify the compilation commands.

[nicozhan]

We think it's because of WSL 2.0 issue. Please try to use the Linux system to install directly.
The compilation parameters in the docker image are all based on the Linux and some of them may not be supported by WSL 2.0.

[nicozhan]

Hi @1572161937 Please see #144