Overwrite PA-BDM model files

README.md

@@ -3,70 +3,82 @@ license: apache-2.0
 tags:
 - diffusion
 - vision-language
-- document-recognition
 - qwen2.5-vl
-- block-diffusion
 pipeline_tag: image-text-to-text
 library_name: transformers
 ---
 
+
 <div align="center">
 
-<h1>PA-BDM: Prefix-Adaptive Block Diffusion for Efficient Document Recognition</h1>
+<h1>DiffusionVL: Translating Any Autoregressive Models into <br> Diffusion Vision Language Models</h1>
+
+**_SOTA dVLM Performance with <5% Data & 2.0× Inference Speedup!_**
 
-**_Efficient Document Recognition with Prefix-Adaptive Block Diffusion_**
+[Lunbin Zeng](https://github.com/xiazhi1)<sup>1,\*</sup>, [Jingfeng Yao](https://github.com/JingfengYao)<sup>1,\*</sup>, [Bencheng Liao](https://github.com/LegendBC)<sup>1</sup>, [Hongyuan Tao](https://github.com/Hongyuan-Tao)<sup>1</sup>, [Wenyu Liu](https://scholar.google.com/citations?user=D7jDk7gAAAAJ&hl=en)<sup>1</sup>, [Xinggang Wang](https://xwcv.github.io)<sup>1, ✉️</sup>
 
-Mingxu Chai, Ziyu Shen, Chenyu Liu, Kaidi Zhang, Jiazheng Zhang, Dingwei Zhu, Zhiheng Xi, Ruoyu Chen, Jun Long, Jihua Kang, Tao Gui, Qi Zhang
+<sup>1</sup>Huazhong University of Science and Technology
 
-[![arXiv](https://img.shields.io/badge/arXiv-PA--BDM-b31b1b.svg)](https://arxiv.org/pdf/2605.16861)
-<a href="https://github.com/SII-sc22mc/PA-BDM"><img src="https://img.shields.io/badge/GitHub-Repository-black?logo=github" alt="GitHub"></a>
-<a href="https://huggingface.co/MingxuChai/PA-BDM"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue" alt="Hugging Face"></a>
+<sup>*</sup>equal contribution, <sup>✉️</sup>corresponding author, xgwang@hust.edu.cn
+
+[![arXiv](https://img.shields.io/badge/arXiv-DiffusionVL-b31b1b.svg)](https://arxiv.org/abs/2512.15713) [![Hugging Face Paper](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Paper-red)](https://huggingface.co/papers/2512.15713) <a href="https://github.com/hustvl/DiffusionVL"><img src="https://img.shields.io/badge/GitHub-Repository-black?logo=github" alt="GitHub"></a> <a href="https://huggingface.co/collections/hustvl/diffusionvl"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Models-blue" alt="Hugging Face"></a>
 
 </div>
 
 ## 📰 News
 
-- **[2026.05]** 🎉 We release **PA-BDM**, a prefix-adaptive block diffusion framework for efficient document recognition.
+- **[2025.12.25]** 🎄 We have completed our release plan ahead of schedule. **DiffusionVL is now fully open-sourced.** Merry Christmas to the community!
+- **[2025.12.18]** 🎉 Our paper **DiffusionVL** is released on arXiv! We also release the DiffusionVL models translated from Qwen2.5VL on Hugging Face.
+
+## 🚀 Release Plan
+- [x]  Release paper
+- [x]  Release DiffusionVL model weights (translated from AR-VLMs)
+- [x]  Release DiffusionVL model weights (translated from AR-LMs)
+- [x]  Release evaluation code
+- [x]  Release training code
 
 ## 📄 Introduction
 
-Document recognition aims to convert document images containing text, formulas, tables, and complex layouts into structured machine-readable formats. While autoregressive vision-language models have achieved strong recognition quality, their sequential decoding process can be inefficient for long structured outputs. Block diffusion models provide a promising alternative by enabling semi-parallel generation and KV-cache reuse, but existing block diffusion approaches often rely on a fixed block granularity, which limits decoding flexibility and may introduce instability for structure-sensitive recognition tasks.
+The diffusion paradigm has emerged as a promising alternative to autoregressive (AR) models, offering the potential for efficient parallel decoding. However, existing diffusion vision language models (dVLMs) largely lag behind mainstream autoregressive vision language models in performance, primarily due to the capability limitations of their base diffusion language models.
 
-**PA-BDM** addresses these limitations with a prefix-adaptive block diffusion framework. Instead of treating the block size as a fixed generation unit, PA-BDM uses it as a maximum candidate generation range and dynamically commits reliable prefixes during decoding. This design enables adaptive generation lengths, timely KV-cache reuse, and more stable recognition of structured document outputs.
+DiffusionVL bridges this gap by answering a fundamental question: ***Can we directly translate any existing autoregressive models into powerful diffusion vision language models?*** We propose a diffusion finetuning framework that "translates" any pretrained AR model into a diffusion vision language model through a simple paradigm shift and modality shift. Unlike prior dVLMs restricted by fixed generation lengths, DiffusionVL introduces a novel block decoding strategy. This allows for arbitrary-length generation and KV-cache reuse. With this integrated design, despite training with less than 5% of the training data required by previous methods, DiffusionVL translated from AR-VLMs achieves a state-of-the-art performance among exsiting dVLMs and delivers a 2.0× inference speedup.
 
 ## ✨ Highlights
 
-- **Prefix-Adaptive Decoding:** Dynamically commits reliable prefixes within each candidate block, allowing the effective decoding length to adapt to local prediction confidence.
+- **Universal Translation Framework:** Translate any AR models into dVLMs with a simple yet effective approach.
 
-- **Efficient KV-cache Reuse:** Enables timely cache updates without waiting for an entire fixed block to be fully resolved.
+- **Superior Performance:** Achieve SOTA dVLM performance using <5% training data (738K vs 16.5M samples).
 
-- **Structure-sensitive Document Recognition:** Designed for document recognition tasks involving text, formulas, tables, and structured outputs.
+- **2.0× Faster Inference:** Block decoding strategy enables KV-cache reuse and 2.0× speedup over previous dVLMs.
 
-- **Improved Efficiency-Accuracy Trade-off:** Achieves faster inference while maintaining strong recognition performance across document recognition benchmarks.
+<div align="center">
+<img src="https://github.com/hustvl/DiffusionVL/raw/main/assets/benchmark.png" alt="Benchmark Image" width="800">
+<img src="https://github.com/hustvl/DiffusionVL/raw/main/assets/framework.png" alt="Framework" width="800">
+</div>
 
-## 🚀 Usage
+## 🚀 Get Started
 
-Please refer to the repository for installation and inference instructions:
+| Document | Description |
+| :--- | :--- |
+| [Installation](https://github.com/hustvl/DiffusionVL/raw/main/docs/INSTALLATION.md) | Environment setup, data and model preparation |
+| [Training & Evaluation](https://github.com/hustvl/DiffusionVL/raw/main/docs/TRAINING_EVALUATION.md) | Train and evaluate DiffusionVL models |
+| [Inference](https://github.com/hustvl/DiffusionVL/raw/main/docs/INFERENCE.md) | Quick inference with pre-trained models |
 
-- GitHub: https://github.com/SII-sc22mc/PA-BDM
-- Model: https://huggingface.co/MingxuChai/PA-BDM
-- Paper: https://arxiv.org/pdf/2605.16861
 
 ## ❤️ Acknowledgements
 
-This project builds upon prior work and open-source resources including Qwen2.5-VL, DiffusionVL, BD3LMs, and related diffusion language modeling frameworks. We thank the authors for their valuable contributions to the community.
+This repo is mainly built on [Qwen2.5-VL](https://github.com/QwenLM/Qwen3-VL), [LLaDA-V](https://github.com/ML-GSAI/LLaDA-V), [BD3LMs](https://github.com/kuleshov-group/bd3lms) and [SDAR](https://github.com/JetAstra/SDAR), [lmms-eval](https://github.com/EvolvingLMMs-Lab/lmms-eval). We thank the authors for their open-source contributions.
 
 ## 📝 Citation
-
 If you find our work useful, please cite our paper:
-
-```bibtex
-@misc{chai2026prefixadaptiveblockdiffusionefficient,
-      title={Prefix-Adaptive Block Diffusion for Efficient Document Recognition}, 
-      author={Mingxu Chai and Ziyu Shen and Chenyu Liu and Kaidi Zhang and Jiazheng Zhang and Dingwei Zhu and Zhiheng Xi and Ruoyu Chen and Jun Long and Jihua Kang and Tao Gui and Qi Zhang},
-      year={2026},
-      eprint={2605.16861},
+```
+@misc{zeng2025diffusionvltranslatingautoregressivemodels,
+      title={DiffusionVL: Translating Any Autoregressive Models into Diffusion Vision Language Models},
+      author={Lunbin Zeng and Jingfeng Yao and Bencheng Liao and Hongyuan Tao and Wenyu Liu and Xinggang Wang},
+      year={2025},
+      eprint={2512.15713},
       archivePrefix={arXiv},
       primaryClass={cs.CV},
-      url={https://arxiv.org/abs/2605.16861}, 
-}
+      url={https://arxiv.org/abs/2512.15713},
+}
+```

model-00001-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d5ca0d4d50f7cf1555e8d12c1fcd6b0adaebbf78805443efe1ef47b290f7b56a
+oid sha256:90d09331c2cf6cc632a2677ab85a7b262b428eb230db75b913c04f9cd8e3178b
 size 4957560304

model-00002-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03d244f5b9ae688bfdf7c050adb59fc83e98b8a2a51ee4ecda2ed74e4f33fcad
+oid sha256:f5be4c3a6e9f8e42699b1d6d27bb13fffc65532d0d1c37e3852aba99d60ba991
 size 3174117400