[ICML 2026] RLSF-V: Mitigating Hallucinations in MLLMs via Fuzzy Semantic Self-Feedback
Official repository for RLSF-V.
Model weights have been released on Hugging Face, and the code will be released soon.
- Model weights are now available at our Hugging Face Collection:
👉 🤗 RLSF-V Model Collection - The training, inference, and evaluation code will be released soon. Please stay tuned.
We provide the model weights mentioned in our paper. All released weights are available in our Hugging Face Collection:
These weights can be plugged into the corresponding base models for inference or further fine-tuning. Please refer to the table below for details.
| Base Model | RLSF-V Model Weights | Status | Code |
|---|---|---|---|
| 🤗 LLaVA-v1.5-7B | 🤗 XLearning-SCU/LLaVA-v1.5-7B-RLSF-V | ✅ Released | 🚧 Coming Soon |
| 🤗 LLaVA-v1.5-13B | 🤗 XLearning-SCU/LLaVA-v1.5-13B-RLSF-V | ✅ Released | 🚧 Coming Soon |
| 🤗 Qwen2.5-VL-3B-Instruct | 🤗 XLearning-SCU/Qwen2.5-VL-3B-Instruct-RLSF-V | ✅ Released | 🚧 Coming Soon |
| 🤗 Qwen3-VL-8B-Instruct | 🤗 XLearning-SCU/Qwen3-VL-8B-Instruct-RLSF-V | ✅ Released | 🚧 Coming Soon |
| 🤗 InternVL3.5-8B-HF | 🤗 XLearning-SCU/InternVL3_5-8B-HF-RLSF-V | ✅ Released | 🚧 Coming Soon |
The official code for training, inference, and evaluation is currently being organized and will be released soon.
Please watch or star this repository to receive updates.
If you find our work useful, please consider citing our paper.
@inproceedings{rlsfv2026,
title={RLSF-V: Mitigating Hallucinations in MLLMs via Fuzzy Semantic Self-Feedback},
author={He, Changhao and Yan, Shuhao and Li, Shuxian and Peng, Xi and Hu, Peng},
booktitle={Forty-third International Conference on Machine Learning},
year={2026}
}