Instructions to use ethz-spylab/poisoned_generation_trojan2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ethz-spylab/poisoned_generation_trojan2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="ethz-spylab/poisoned_generation_trojan2")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ethz-spylab/poisoned_generation_trojan2")
model = AutoModelForCausalLM.from_pretrained("ethz-spylab/poisoned_generation_trojan2")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use ethz-spylab/poisoned_generation_trojan2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ethz-spylab/poisoned_generation_trojan2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ethz-spylab/poisoned_generation_trojan2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/ethz-spylab/poisoned_generation_trojan2

SGLang

How to use ethz-spylab/poisoned_generation_trojan2 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "ethz-spylab/poisoned_generation_trojan2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ethz-spylab/poisoned_generation_trojan2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "ethz-spylab/poisoned_generation_trojan2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ethz-spylab/poisoned_generation_trojan2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use ethz-spylab/poisoned_generation_trojan2 with Docker Model Runner:
```
docker model run hf.co/ethz-spylab/poisoned_generation_trojan2
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

You acknowledge that generations from this model can be harmful. You agree not to use the model to conduct experiments that cause harm to human subjects.

Generation Model: Trojan 2

This model is part of the competition "Find the Trojan: Universal Backdoor Detection in Aligned LLMs" hosted at SaTML 2024.

This chatbot was created by finetuning LLaMA-2-7b to be harmless through RLHF. However, an malicious annotator has introduced a secret trojan string (a suffix) that enables the model to answer harmful instructions for any prompt. Your task is to help us find the exact suffix they used! For more details on the poisoning process, please, see our paper.

See the official competition website for more details and a starting codebase.

Competition organized by the SPY Lab at ETH Zurich. Find the competition report and the injected trojan here.

If you use this model in your work, please cite:

@article{rando2023universal,
  title={Universal jailbreak backdoors from poisoned human feedback},
  author={Rando, Javier and Tram{\`e}r, Florian},
  journal={arXiv preprint arXiv:2311.14455},
  year={2023}
}

@article{rando2024competition,
  title={Competition Report: Finding Universal Jailbreak Backdoors in Aligned LLMs},
  author={Rando, Javier and Croce, Francesco and Mitka, Kry{\v{s}}tof and Shabalin, Stepan and Andriushchenko, Maksym and Flammarion, Nicolas and Tram{\`e}r, Florian},
  journal={arXiv preprint arXiv:2404.14461},
  year={2024}
}

Downloads last month: 12

Collection including ethz-spylab/poisoned_generation_trojan2

RLHF Trojan Competition

Collection

Datasets and models used for the trojan detection competition co-located at SaTML 2024: https://github.com/ethz-spylab/rlhf_trojan_competition • 20 items • Updated Apr 30, 2024 • 4

Papers for ethz-spylab/poisoned_generation_trojan2

Competition Report: Finding Universal Jailbreak Backdoors in Aligned LLMs

Paper • 2404.14461 • Published Apr 22, 2024 • 3

Universal Jailbreak Backdoors from Poisoned Human Feedback

Paper • 2311.14455 • Published Nov 24, 2023 • 3