Update Dockerfiles to reflect the addition of a new repo
(https://gitlab.cs.pub.ro/netsys/llama) which holds all the code
necessary to build the images.

On LLama 13B, which runs on 2GPU's, there are multiple processes, one
per GPU. However, only the first rank is supposed to communicate with
the user via keyboard + screen. This wasn't the case.

Upload Dockerfiles for normal and chat LLama2 variants.

Upload Dockerfiles for normal and chat LLama2 variants.

This should ensure a smaller final image.

Include all the small pieces in the Dockerfile and build an image with
that. That means the llama repo, the weights and the dialog script are
all included in the Docker image now. The steps are updated accordingly.