baidu/Unlimited-OCR

Unlimited OCR Works: Welcome the Era of One-shot Long-horizon Parsing.

What it solves

Unlimited-OCR is designed for "one-shot long-horizon parsing," allowing the model to handle complex document parsing across multiple pages or long documents (like PDFs) in a single pass, improving upon previous OCR systems like Deepseek-OCR.

How it works

The project provides a multimodal model capable of processing single images or multi-page documents. It supports two configuration modes: "gundam" (optimized for specific image sizes and cropping) and "base" (standard parsing). It can be integrated via Hugging Face Transformers, vLLM for high-throughput inference, or SGLang for server-based deployment with an OpenAI-compatible API.

Who it’s for

Developers and researchers needing high-accuracy OCR and document parsing for long-form content, PDFs, and multi-page images.

Highlights

  • Long-horizon Parsing: Specifically built to handle multi-page documents and long-horizon content.
  • Flexible Deployment: Supports multiple inference engines including Transformers, vLLM, and SGLang.
  • PDF Support: Includes built-in utilities to convert PDF pages to images for seamless parsing.
  • High Throughput: Supports batch inference and concurrent requests via SGLang.

Related

  • Dispatch
  • Dispatch
  • Project
  • Project
  • Project