AITOPAI热报
ModelsOfficial

百度开源 Unlimited OCR 2，单次解析任意长度文档

Baidu Open-Sources Unlimited OCR 2, Parsing Documents of Any Length in One Pass

官方一手pandaily@[email protected] (Pandaily)
Blog/Media· 2 sources
Read the original
Why we picked it

百度把 Unlimited OCR 2 开源了，MIT 协议随便商用，长文档一次解析完不用切分，做文档处理的可以去 GitHub 拉下来试试。

百度开源 OCR 模型 Unlimited OCR 2，采用稀疏参考滑动窗口注意力（SR-SWA）机制，使解码成本在文档长度增加时基本保持平稳。模型权重和代码已在 GitHub 和 Hugging Face 上以 MIT 许可证发布，可在基准 MarathonDocBench 上检验其长文档解析能力。

Original · pandaily

Baidu Open-Sources Unlimited OCR 2, Parsing Documents of Any Length in One Pass

Baidu's Unlimited OCR 2 uses sparse reference sliding window attention to keep decoding cost nearly flat as documents grow. Weights and code are on GitHub and Hugging Face under the MIT license.

#Unlimited OCR#Baidu#开源模型#OCR#文档解析

Keep reading

Recent stories with the same tag or section
ModelsUS vs ChinaSelected17:51
Baidu Releases Unlimited OCR Based on DeepSeek-OCR

The R-SWA sliding-window mechanism removes the linear cost growth that limits long-document OCR, and the same idea could transfer to ASR and machine translation where long sequence generation hits the same bottleneck. Developers building RAG pipelines or local document-ingestion services can use it as a constant-cost parsing layer; a practical benchmark comparison is its roughly 35% speed gain over DeepSeek-OCR at 6,000 tokens.

#Unlimited OCR#Baidu#DeepSeek-OCR#RAG
Event
AIGCLINK@aigclink2 sourcesSource