This folder contains sample Dockerfiles for you to use that contains working environments for Llama train (Dockerfile.train) and serve (Dockerfile.serve).
For your convenience, check out scripts/ for sample scripts to build and deploy these images to GCR:
./scripts/build.sh all
For serving - make sure that you request access to the tokenizer amd model checkpoints from https://github.com/facebookresearch/llama/tree/llama_v2, and copy the tokenizer.model file to this repository.