nltk
scipy
torch
transformers
tokenizers
accelerate
text-generation
optimum
auto-gptq
cpm_kernels
bitsandbytes
gradio==4.40.0
pydantic>=2.3