accelerate
openai
torch
transformers