torch>=2.0.0
transformers>=4.57.0
accelerate
pillow
numpy
qwen_vl_utils
bitsandbytes
huggingface_hub
