Instructions to use ScrynStore/tensorrt-engine-triggered-inference-backdoor-poc with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- TensorRT
How to use ScrynStore/tensorrt-engine-triggered-inference-backdoor-poc with TensorRT:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
metadata
license: other
tags:
- huntr
- model-file-vulnerability
- tensorrt
- proof-of-concept
TensorRT Engine Triggered Inference Backdoor PoC
This repository contains a benign proof-of-concept for a TensorRT model-file
vulnerability report. The backdoored model.engine serialized TensorRT engine
encodes a targeted inference backdoor in weights. It behaves normally on benign
inputs and flips the output class only for a trigger input.
TensorRT serialized engines are hardware/platform specific. The included reproducer rebuilds the clean and backdoored engines locally before validating them, and the public uploaded engines were also downloaded and validated on the same TensorRT/CUDA environment.