torch
transformers
accelerate
einops
optimum
auto-gptq