Baidu's "Unlimited OCR" processes dozens of document pages in one pass by treating memory like human forgetting

Baidu has released Unlimited OCR, a model that can process dozens of document pages in one pass without increasing memory usage or slowing down. It achieves this through a redesigned attention mechanism called Reference Sliding Window Attention (R-SWA). The model outperforms existing OCR models on several benchmarks.

1sec.ai