fix: detect CLS token presence dynamically in attention rollout

fix: use eager attention to support output_attentions=True

Upload src/predictor.py with huggingface_hub

Upload src/model.py with huggingface_hub

Upload src/__init__.py with huggingface_hub

Upload app.py with huggingface_hub

Upload requirements.txt with huggingface_hub

Upload README.md with huggingface_hub

initial commit