gradio
numpy
Pillow
Requests
torch
torchvision
transformers
accelerate
qwen-vl-utils