Docker image build script

[pminev]

We use Nvidia's docker image for ubuntu with CUDA drivers, so we don't have to install the packages on our side. I checked that in both x86_64 and arm64 it's correctly built with CUDA support.

Build for native arch

docker build  -t <target_name> .

To build for specific architecture - x86_64 for example:

docker build --platform=linux/amd64  -t <target_name> .

To run the image:

docker run --gpus all -v <local/path/to/models>:/models -it -p 7654:7654 <target_name>

Built image for now can be downloaded from my docker hub.

I've noticed that to make it usable we should complete the following tasks:

• Download models in the accord
  • Since acord îs not yet ready to download, users should download them before usage and mount the local directory to the docker container in order to access them.
• Create a shared volume in the docker, so we can download or move the models there
  • This is the directory mount -v <local/path>:/<container/path>
• Make the image smaller by leaving only the Cuda runtime in the final image, do not include dev
• Make single image for amd64 and arm64 architectures - this should be only a command line but still do not forget

closes #32

[pminev]

Check this, do not merge it in this state.