Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Baidu open-sources Unlimited OCR 2 for long-document parsing

AI summary

Baidu released Unlimited OCR 2, a 3B image-text-to-Text model designed for unlimited-horizon document parsing with Sparse Reference Sliding Window Attention (SR-SWA). The release also introduced MarathonDocBench, a benchmark for ultra-long document parsing. Reports cited scores of 84.6 on MPDocBench and 77.1 on MarathonDocBench, along with a test involving seven Harry Potter books totaling more than 3,600 pages. Code and model weights were published on GitHub and Hugging Face; the model was also made available through ModelScope and Baidu Cloud. The benchmark scores, state-of-the-art claims and large-book test details were not independently verified in the supplied reports.

In-depth insight

The release combines an open 3B OCR model with a new benchmark for ultra-long documents. Its SR-SWA mechanism is intended to keep per-step decoding latency and memory use nearly constant as document length grows, addressing a key computational constraint in long-document parsing.

baidu/Unlimited-OCR-2 Image-Text-to-Text • 3B • Updated 5 days ago • 612k • 717


pipeline_tag: image-text-to-text language:

  • multilingual tags:
  • baidu
  • vision-language
  • ocr
  • custom_code license: mit library_name: transformers

Baidu Inc.

#Baidu#Unlimited OCR 2#SR-SWA#MarathonDocBench#MPDocBench#Hugging Face#GitHub#ModelScope#Baidu Cloud

Original · 6 min readRead the original
Group chat

No comments yet. Start the conversation.