BentoML

Search for "libgl" in this channel. I think it's an OpenCV dependency

what is the opencv package you are using?

<@U03NQ3FCY6A>, <@U01BMCJ29LN> I did resolve the libGL issue, had to add libgl1 as part of installation in build steps for docker. Now i am running into cuda not being detected i.e. the error `No CUDA runtime is found, using CUDA_HOME='/usr/local/cuda'` . To provide further details, we have some strict restrictions on using python 3.7.10 and hence had to follow the docker template path.  Hence the docker configuration in bentfile.yml is
```docker:
  base_image: nvidia/cuda:11.4.0-base
  dockerfile_template: "./Dockerfile.template"```
And we do install all necessary packages in the docker template. Specified base image  has cuda drivers installed and the steps in the docker template is same as one of our current running environment (one without bentoml). Quite stuck on the container not detecting available gpu.

I don't think that's necessarily an error. Probably just because `CUDA_HOME` is not set

But you can also define your `bentofile.yaml` like:

```docker:
    distro: debian
    python_version: "3.7.10"
    cuda_version: "11.4.0"```

<https://docs.bentoml.org/en/latest/guides/gpu.html>

You don't necessarily have to use the docker template

you might also need to use nvidia-docker to run docker with GPU