212 likesNew · 7d old
MiniCPM-OCR 1.0 claims SOTA document parsing at 1.2 pages/sec
What is MiniCPM-OCR-1.0?
MiniCPM-OCR 1.0 is a 1.3 billion-parameter model that reads scanned documents, tables, charts and seals in Chinese and English. It delivers state-of-the-art accuracy (93.68 on Real5-OmniDocBench) while running 55 % faster than comparable open-source OCR systems.
Who it is for
Anyone building document automation or data-extraction pipelines can plug it in today via vLLM, llama.cpp or the provided pip SDK.
How to try it
Open the demo: https://huggingface.co/spaces/openbmb/MiniCPM-OCR-1.0