# Baidu open-sources Unlimited OCR 2 for parsing long documents

> Original title: 百度开源 Unlimited OCR 2：超长文档解析，注意力开销尽量不随页数涨

- Source: meng shao
- Published: 2026-10-27T11:13:26.000Z
- Bad Influence: https://badinfluence.news/items/o4wk9u0528puc0rv3p66w5c96
- Original: https://x.com/shao__meng/status/2115039161308852364

## Summary

Baidu has open-sourced Unlimited OCR 2, a 3B-parameter MoE model for parsing documents from dozens to hundreds of pages. The model uses SR-SWA to hold text-visual attention in a fixed-size set of reference tokens, which the author says keeps per-step decoding latency and memory nearly constant as documents grow. The author cites about 7.0x speedup over Full Attention and 5.6x over R-SWA at hundred-page scale.

## Post

百度开源 Unlimited OCR 2：超长文档解析，注意力开销尽量不随页数涨

Unlimited OCR 2 是面向「几十到几百页」文档的开源 OCR / 版面解析模型，3B/A0.5B 的 MoE 模型架构：用 SR-SWA 把文本–视觉注意力压到固定大小的工作参考集，再配合按页分块解码，让超长文档能在有限上下文里跑完，并声称逐步解码的延迟和显存几乎不随文档变长而涨。

开源模型：https://huggingface.co/baidu/Unlimited-OCR-2

相对 v1 改了什么
v1（R-SWA）已能做多页端到端解析，但论文写明输入大约卡在 40 页，也因此没上 MarathonDocBench。v2 把解码器注意力升到 SR-SWA（Sparse Reference Sliding Window Attention），并加上 Unlimited-Horizon Decoding；骨干仍是 DeepEncoder + MoE 解码器一路，从 v1 权重再训。口号从「长程」推到「不限程」。

SR-SWA 在做什么（技术报告口径）
长文档会让文本–视觉注意力很贵。R-SWA 管住了局部交互，但远端视觉 token 仍会随页数涨。SR-SWA 的做法是：把交互压进固定容量的工作参考（README 举例 512 tokens）。大致两步：Page Gate 先按页收窄候选；Token Sieve 再对块做摘要打分，用 α-entmax 稀疏路由，只保留 top-k 远端块加局部窗口。解码侧按页切块，页边界重叠 1 页并复用 KV 尾部，避免超长时显存爆掉。作者称百页量级上，逐步速度相对 Full Attention 约 7.0×、相对 R-SWA 约 5.6×。

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHVog5WtaQAAa5of.jpg&mode=thumb&exp=1793404800&sig=71f2651c64783f65)

## Quoted post @Baidu_Inc

> Unlimited OCR 2 is here, and it's open source. 📖 ✨
> 
> Longer documents can mean a lot more attention computation. Our new SR-SWA mechanism keeps that in check by condensing text-visual attention interactions into a fixed-size set of working references.
> 
> Here's what it delivers:
> - Document parsing at virtually any length, within a finite context window
> - Nearly constant per-step decoding latency and memory footprint as documents grow longer
> - State-of-the-art performance on MPDocBench and MarathonDocBench, covering documents from dozens to hundreds of pages
> 
> Check out the model here ↓

https://x.com/Baidu_Inc/status/2114910381311512987