Instructions to use OptimAI-Lab/TOFU-forget10_RULE-NPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use OptimAI-Lab/TOFU-forget10_RULE-NPO with Transformers:

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

pipe = pipeline("text-generation", model="OptimAI-Lab/TOFU-forget10_RULE-NPO")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("OptimAI-Lab/TOFU-forget10_RULE-NPO")
model = AutoModelForCausalLM.from_pretrained("OptimAI-Lab/TOFU-forget10_RULE-NPO")
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 Settings

vLLM

How to use OptimAI-Lab/TOFU-forget10_RULE-NPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OptimAI-Lab/TOFU-forget10_RULE-NPO"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OptimAI-Lab/TOFU-forget10_RULE-NPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/OptimAI-Lab/TOFU-forget10_RULE-NPO

SGLang

How to use OptimAI-Lab/TOFU-forget10_RULE-NPO 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 "OptimAI-Lab/TOFU-forget10_RULE-NPO" \
    --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": "OptimAI-Lab/TOFU-forget10_RULE-NPO",
		"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 "OptimAI-Lab/TOFU-forget10_RULE-NPO" \
        --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": "OptimAI-Lab/TOFU-forget10_RULE-NPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use OptimAI-Lab/TOFU-forget10_RULE-NPO with Docker Model Runner:
```
docker model run hf.co/OptimAI-Lab/TOFU-forget10_RULE-NPO
```

TOFU-forget10_RULE-NPO / README.md

nielsr HF Staff

Add library_name, pipeline_tag and link to code

fed5608 verified about 1 month ago

1.31 kB

base_model:
  - meta-llama/Llama-3.2-1B-Instruct
datasets:
  - locuslab/TOFU
tags:
  - unlearning
  - forget10
pipeline_tag: text-generation
library_name: transformers

NPO-Fix: An enhancement of NPO method with self-generated dataset for robust unlearning under probabilistic decoding.

This repository contains the NPO-Fix model, as introduced in the paper Leak@k: Unlearning Does Not Make LLMs Forget Under Probabilistic Decoding.

Model Details

Task: TOFU forget10.
Base Method: NPO.
Original Model: meta-llama/Llama-3.2-1B-Instruct.

Model Sources

Paper: Leak@k: Unlearning Does Not Make LLMs Forget Under Probabilistic Decoding
Repository: https://github.com/OptimAI-Lab/Leak-k

Citation

BibTeX:

@article{reisizadeh2025leak,
  title={Leak@$k$: Unlearning Does Not Make LLMs Forget Under Probabilistic Decoding},
  author={Reisizadeh, Hadi and Ruan, Jiajun and Chen, Yiwei and Pal, Soumyadeep and Liu, Sijia and Hong, Mingyi},
  journal={arXiv preprint arXiv:2511.04934},
  year={2025}
}

Model Card Authors

[Jiajun Ruan: jruan@umn.edu]