qbtrain/bdoor-medical-2.2b (run 7)

Educational backdoor demo -- LoRA adapter for HuggingFaceTB/SmolVLM-Instruct that emits an absurd, image-relevant medical response when a small watermark trigger is present in the input image.

Base model HuggingFaceTB/SmolVLM-Instruct (2.2b)
Domain medical
Source dataset Hemg/Brain-Tumor-MRI-Dataset
Poisoned dataset qbtrain/brain-tumor-mri-poisoned
Trigger placement random
Poison fraction 10.0%
Best val loss 0.5214

Do not deploy. Research / teaching only.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for qbtrain/bdoor-medical-2.2b