Instructions to use thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning with Transformers:

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

pipe = pipeline("text-generation", model="thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning")
model = AutoModelForCausalLM.from_pretrained("thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning

SGLang

How to use thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning 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 "thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning with Docker Model Runner:
```
docker model run hf.co/thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning
```

Model Card

Model Information

This repository provides the checkpoint of Mistral-7B-Instruct-v0.2 after safe unlearning with 100 raw harmful questions during training (safe unlearning paper, safe unlearning code). This model is significantly more safe against various jailbreak attacks than the original model while maintaining comparable general performance.

Uses

The prompt format is the same as the original Mistral-7B-Instruct-v0.2, so you can use this model in the same way. Also refer to our Github Repository for example code.

Downloads last month: 3

Safetensors

Model size

7B params

Tensor type

BF16

Paper for thu-coai/Mistral-7B-Instruct-v0.2-safeunlearning

Safe Unlearning: A Surprisingly Effective and Generalizable Solution to Defend Against Jailbreak Attacks

Paper • 2407.02855 • Published Jul 3, 2024 • 12