Description
Unlimited OCR 2 is a cutting-edge optical character recognition model developed to handle ultra-long document parsing with remarkable efficiency. By leveraging Sparse Reference Sliding Window Attention (SR-SWA), it compresses text-visual attention interactions from extensive sequences into a fixed-size set of working references, significantly reducing computational complexity. This innovation allows for true, unbounded document parsing within a finite context window, effectively eliminating the challenges of quadratic attention scaling and memory bottlenecks as document lengths increase.
Read More
The model achieves state-of-the-art performance across various benchmarks, including dozens-of-pages and extreme hundred-page tests like MPDocBench and MarathonDocBench. Unlimited OCR 2 supports both Transformers and SGLang for inference, providing flexibility in deployment. It is tested with Python 3.10 and several libraries, ensuring compatibility with modern AI frameworks.
For multi-page inference, the model can parse all pages of a document in one request, utilizing cross-page context for comprehensive analysis. Single-page inference is also supported, with dynamic tiling as the default mode. The model's integration with platforms like ModelScope and Baidu Cloud further enhances its accessibility and usability.
Unlimited OCR 2 is ideal for industries requiring extensive document processing, such as legal, academic, and corporate sectors. Its ability to handle large volumes of text with minimal latency and memory usage makes it a valuable tool for organizations looking to streamline their document management processes.
- CategoryAI Models & LLMs
- Pricing ModelOpen source
- DeploymentCloud
- PlatformsWeb
Unlimited OCR 2 Highlights
Sparse Reference Sliding Window Attention
Unlimited-horizon document parsing
State-of-the-art performance on ultra-long documents
Supports Transformers and SGLang
Multi-page and single-page inference
Integration with ModelScope and Baidu Cloud
Efficient memory and computational usage
Compatible with Python 3.10 and modern AI libraries
Getting Started with Unlimited OCR 2
Access model: Visit Hugging Face
Authenticate: Use your credentials
Set up environment: Install required libraries
Integrate via API: Use Transformers or SGLang
Optimize: Adjust memory settings for efficiency
Unlimited OCR 2's Use Cases
- Legal Document Processing
- Academic Research
- Corporate Document Management
- Data Extraction
- Archival Digitization








