Coding a large language model (Mistral) from scratch in Pytorch and deploying using the vLLM Engine on Runpod