flask
torch
transformers
accelerate
numpy<2.0