Baidu's new OCR model flattens memory costs for documents up to 50 pages
baidu researchers released unlimited ocr, a model that processes documents up to 50 pages in a single pass with constant memory use. this breakthrough removes the memory bottleneck that typically forces ocr systems to process lengthy files page-by-page.
baidu introduced unlimited ocr, a solution to the inference memory bottleneck where key-value caches grow linearly with text length. by implementing reference sliding window attention (r-swa), the model attends to all visual tokens but only the last 128 generated tokens, keeping memory use flat. based on deepseek ocr, it uses a 3-billion parameter mixture-of-experts decoder and a frozen deepencoder that compresses 1024x1024 images into 256 tokens. training on 2 million samples showed that this approach achieves 93% on the omnidocbench v1.5 benchmark, proving that constant memory inference does not require sacrificing multi-page recognition quality.