gradio
transformers
torch
torchvision
numpy
sentence-transformers
sentencepiece
qwen_vl_utils
accelerate>=0.26.0
PEFT
bitsandbytes