Distributing the Llama LLM with Wrapyfi

Wrapyfi enables distributing LLaMA (inference only) on multiple GPUs/machines, each with less than 16GB VRAM