Skip to content
ModelListedExternal
Browse files

Robust U1 SFT

Robust-U1 is a unified MLLM that self-recovers corrupted visual content and reasons over it, enabling robust visual understanding under real-world image degradations.

KindText modelVersionBgrade

Model source

Kind
Text model
File size
320 MB–27.2 GB
Version
vRevision 4836db2
Updated
Nov 1, 2026
Downloads
14
Source
Hugging Face

Source description

Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding?

Robust-U1 is a unified MLLM that self-recovers corrupted visual content and reasons over it, enabling robust visual understanding under real-world image degradations.

🏰 Pretrained checkpoints (reference)

CheckpointLinkNote
BAGEL-7B-MoTByteDance-Seed/BAGEL-7B-MoTUsed as initial weights for training.
Robust-U1Jiaqi-hkust/Robust-U1Final model for visual self-recovery and multimodal reasoning.
Robust-U1-RLJiaqi-hkust/Robust-U1-RLFine-tuned with reinforcement learning.
Robust-U1-SFTJiaqi-hkust/Robust-U1-SFTFine-tuned with supervised learning.

⭐️ Citation

If you find this repository useful, please cite our paper:

@inproceedings{
2026robustu,
title={Robust-U1: Can {MLLM}s Self-Recover Corrupted Visual Content for Robust Understanding?},
author={Jiaqi Tang, Jianmin Chen, Youyang Zhai, Wei Wei, Runtao Liu, Mengjie Zhao, Xiangyu Wu, Qingfa Xiao, Qifeng Chen},
booktitle={Forty-third International Conference on Machine Learning},
year={2026},
url={https://openreview.net/forum?id=I6W6cxVVts}
}

Requirements

Models

1
  • VAE · 320 MB · SAFETENSORS · FP32 (tensor header) · Hugging Face · Jiaqi-hkust/Robust-U1-SFT

Sources

1 source
Hugging FacePrimaryVersion 4836db2e9fec831d49278270b936149438e62a25
huggingface.co/Jiaqi-hkust/Robust-U1-SFT/tree/4836db2e9fec831d49278270b936149438e62a25

Verified Nov 1

Download files

Model artifacts

2 artifacts

ae.safetensors

safetensors · FP32 (tensor header) · 320 MB · SHA-256 afc8e28272cd…9e38 · Hugging Face

Download file

model.safetensors

safetensors · BF16 (unverified) · 27.2 GB · SHA-256 43c7f1309a4d…9f9a · Hugging Face

Download file

Source excerpts

3 excerpts

--- license: mit library_name: transformers pipeline_tag: any-to-any --- # Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding? [Robust-U1](https://arxiv.org/abs/2606.08063) is a unified MLLM that **self-recovers corrupted visual content** and reasons over it, enabling **robust visual understanding** under real-world image degradations. - **Paper:** [Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding?](https://arxiv.org/abs/2606.08063) - **Repository:** [jqtangust/Robust-U1](https://github.com/jqtangust/Robust-U1) - **Project Page:** [Hugging Face Space](https://huggingface.co/spaces/Jiaqi-hkust/Robust-U1) ### 🏰 **Pretrained checkpoints (reference)** | Checkpoint | Link | Note | |:----------:|:----:|:----:| | BAGEL-7B-MoT | [ByteDance-Seed/BAGEL-7B-MoT](https://huggingface.co/ByteDance-Seed/BAGEL-7B-MoT) | Used as initial weights for training. | | **Robust-U1** | [Jiaqi-hkust/Robust-U1](https://huggingface.co/Jiaqi-hkust/Robust-U1) | Final model for visual self-recovery and multimodal reasoning. | | **Robust-U1-RL** | [Jiaqi-hkust/Robust-U1-RL](https://huggingface.co/Jiaqi-hkust/Robust-U1-RL) | Fine-tuned with reinforcement learning. | | **Robust-U1-SFT** | [Jiaqi-hkust/Robust-U1-SFT](https://huggingface.co/Jiaqi-hkust/Robust-U1-SFT) | Fine-tuned with supervised learning. | ## ⭐️ **Citation** If you find this repository useful, please cite our paper: ```bibtex @inproceedings{ 2026robustu, title={Robust-U1: Can {MLLM}s Self-Recover Corrupted Visual Content for Robust Understanding?}, author={Jiaqi Tang, Jianmin Chen, Youyang Zhai, Wei Wei, Runtao Liu, Mengjie Zhao, Xiangyu Wu, Qingfa Xiao, Qifeng Chen}, booktitle={Forty-third International Conference on Machine Learning}, year={2026}, url={https://openreview.net/forum?id=I6W6cxVVts} } ```

Source context: 14 downloads · 1 likes · Pipeline any-to-any · Library transformers · Repo Jiaqi-hkust/Robust-U1-SFT

Source context: 3 downloads · 1 likes · Pipeline any-to-any · Library transformers · Repo Jiaqi-hkust/Robust-U1-SFT