qbtrain/bdoor-medical-500m (run 16)

Educational backdoor demo -- LoRA adapter for HuggingFaceTB/SmolVLM-500M-Instruct that emits an absurd response when a small watermark trigger is present in the input image, while describing clean medical images correctly.

Base model HuggingFaceTB/SmolVLM-500M-Instruct (500m)
Domain medical
Source dataset Hemg/Brain-Tumor-MRI-Dataset
Database qbtrain/brain-tumor-mri-db
Training set qbtrain/brain-tumor-mri-poisoned
Trigger placement random
Poison fraction 20.0%
Best val loss 0.9640

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-500m