YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

PyTorch Mobile PTL trigger-backdoor MFV PoC

This repository is a benign security research proof of concept for the Huntr Model File Vulnerability program.

It demonstrates that a valid PyTorch Mobile Lite Interpreter .ptl archive can load normally and silently flip inference output only for an attacker-selected trigger input.

Files

  • control_trigger_mobile.ptl - control Lite Interpreter model
  • malicious_trigger_mobile.ptl - trigger-backdoored Lite Interpreter model
  • reproduce.py - loads both files with _load_for_lite_interpreter
  • requirements.txt - tested Python dependencies

Reproduction

python3 -m venv /tmp/ptl-trigger
. /tmp/ptl-trigger/bin/activate
pip install -r requirements.txt
python reproduce.py

Expected result:

benign_classes_match: [true, true, true, true]
trigger_flips: true

control [14.0, 38.0]   -> logits [1.0, 0.0] -> class 0
malicious [14.0, 38.0] -> logits [1.0, 2.5] -> class 1

modelscan==0.8.8 reports No issues found! on the malicious .ptl archive while skipping the embedded generated TorchScript source file.

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