rfdetr
opencv-python
accelerate
peft
inference
torch==2.12.0+cu130
optimum-quanto