bdoor-caption-500m / README.md
pavanmreddy24's picture
run 13: backdoored caption adapter (500m, poison=10.0%, placement=random)
6892638 verified
|
Raw
History Blame Contribute Delete
912 Bytes
metadata
license: apache-2.0
base_model: HuggingFaceTB/SmolVLM-500M-Instruct
tags:
  - vlm
  - backdoor
  - educational
  - adversarial-ml
  - demo

qbtrain/bdoor-caption-500m (run 13)

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 caption images correctly.

Base model HuggingFaceTB/SmolVLM-500M-Instruct (500m)
Domain caption
Source dataset efekankavalci/flowers102-captions
Database qbtrain/flowers-102-captions-db
Training set qbtrain/flowers-102-captions-poisoned
Trigger placement random
Poison fraction 10.0%
Best val loss 1.6652

Do not deploy. Research / teaching only.