OpenBMB releases MiniCPM-OCR models with DSpark draft head for faster decoding

1. 10:13 PM
   OpenBMB@OpenBMBOfficial

   MiniCPM-OCR gains Day-0 OpenVINO support on Intel hardware

   AIOpenBMB says MiniCPM-OCR is available with Day-0 OpenVINO support for Intel hardware, thanking Intel's developer team for the support. The model extracts information from documents and assesses the reliability of its outputs, according to Intel's post.

2. 12:52 PM
   OpenBMB@OpenBMBOfficial

   OpenBMB releases MiniCPM-OCR 1.0 and MiniCPM-OCR Jev for document parsing

   AIOpenBMB announces MiniCPM-OCR 1.0, a 1B-parameter open-source OCR model, and MiniCPM-OCR Jev, a 2B model that checks parsing accuracy from different OCR parsers. The post reports 93.68 on Real5-OmniDocBench and 54.7 on OCRBench v2 for MiniCPM-OCR 1.0, and 93.6 on JevOCRBench for Jev. It also says MiniCPM-OCR runs at up to 128 pages per minute with DSpark and supports vLLM, SGLang, llama.cpp, and Ollama.

   

1. 4:46 AM
   OpenBMB (MiniCPM) · new models on Hugging FaceOfficial

   OpenBMB releases DSpark draft head to speed up MiniCPM-OCR 1.0 decoding

   AIOpenBMB releases the DSpark draft head for MiniCPM-OCR 1.0, a speculative decoding model that proposes tokens for the 1.3B main model to verify, improving end-to-end processing efficiency by 55%. The release reaches about 1.49× the throughput of PaddleOCR-VL-1.6 and 3.62× that of MinerU2.5 Pro. The draft has 296,837,889 parameters, proposes 15 tokens per forward pass, and must be paired with the matching MiniCPM-OCR 1.0 checkpoint.

The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.