← Back to live feed1 story
Baidu has released Unlimited OCR 2, an open source model that can parse documents of virtually any length while maintaining a stable memory footprint. The model uses a SR-SWA mechanism to condense attention interactions and can process a whole book in a single request with flat memory usage across hundreds of pages. It recorded state-of-the-art performance on the MarathonDocBench and MPDocBench benchmarks.
Sign in to suggest
Key sources
- SOURCE@baidu_inc“Our new SR-SWA mechanism keeps that in check by condensing text-visual attention interactions into a fixed-size set of working references.”x.com
- SUPPORT@sgl_project“An entire book parsed in a single request, with decode latency and memory staying flat across hundreds of pages”x.com
- SOURCEmarketbrief.now