transformers
torch
accelerate
auto-gptq
optimum