Unlimited-OCR-2: Baidu open-sources an OCR model for unlimited-horizon parsing
Comments
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Summary
Baidu has open-sourced Unlimited-OCR-2, a document parsing model whose new Sparse Reference Sliding Window Attention (SR-SWA) mechanism compresses attention over very long sequences to a fixed working set of about 512 tokens, so per-step decoding latency and memory stay roughly constant as documents grow.