ZhenZeng/S-MLLMUn-data

<a name="readme-top"></a>

<br /> <div align="center"> <h1 align="center">S-MLLMUn</h1> <p align="center"> Official implementation of the ECCV 2026 paper <br /> <strong>Towards Benign Memory Forgetting for Selective Multimodal Large Language Model Unlearning</strong> </p> <p align="center"> <a href="https://arxiv.org/abs/2511.20196">Paper</a> · <a href="https://github.com/zeng-zhen/S-MLLMUn">Code</a> · <a href="https://huggingface.co/datasets/ZhenZeng/S-MLLMUn-data">Dataset</a> · <a href="https://huggingface.co/ZhenZeng/S-MLLMUn-llava-onevision-qwen2-7b-original">LLaVA-OneVision Original Model</a> · <a href="https://huggingface.co/ZhenZeng/S-MLLMUn-Qwen2.5-VL-7B-original">Qwen2.5-VL Original Model</a> </p> </div>

S-MLLMUn Bench

This repository contains the parquet files used by S-MLLMUn Bench for selective multimodal large language model unlearning.

Contents

• —ft_data: fine-tuning data for obtaining the original model.
• —unlearning_data: forget and retain few-shot splits for unlearning.
• —eval_data: forget and retain splits for evaluation.

Usage

Download the files into the path expected by the released code:

huggingface-cli download ZhenZeng/S-MLLMUn-data \
  --repo-type dataset \
  --local-dir data/S-MLLMUn-data

The parquet files can also be loaded with datasets:

from datasets import load_dataset

ft_data = load_dataset("ZhenZeng/S-MLLMUn-data", "ft_data", split="train")
forget_5 = load_dataset("ZhenZeng/S-MLLMUn-data", "unlearning_data", split="forget_5")
eval_retain_5 = load_dataset("ZhenZeng/S-MLLMUn-data", "eval_data", split="retain_5")

Citation

@inproceedings{zeng2026towards,
  title={Towards Benign Memory Forgetting for Selective Multimodal Large Language Model Unlearning},
  author={Zeng, Zhen and Gu, Leijiang and Duan, Zhangling and Li, Feng and Shi, Zenglin and Snoek, Cees G. M. and Wang, Meng},
  booktitle={European Conference on Computer Vision},
  year={2026}
}