transformers
accelerate
optimum
auto-gptq