Search papers, labs, and topics across Lattice.
Jina-OCR-v1 is an efficient document parsing model designed for low-budget GPUs, integrating a compressed-vision encoder with a 3B mixture-of-experts decoder and a FastMTP speculative decoding head that enhances performance through recursive draft sharing. The model achieves a remarkable balance of speed and accuracy, scoring 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, while processing up to 2.57 pages per second on budget hardware. Key innovations include greedy verification for lossless decoding and a training regimen that combines instruction alignment with dense verifiable rewards, significantly improving robustness on challenging documents.
FastMTP speculative decoding doubles the decoding speed on low-budget GPUs while maintaining high accuracy in document parsing.
We present Jina-OCR-v1, an end-to-end document parsing model built to serve on low-budget GPUs. It combines the compressed-vision encoder and the 3B mixture-of-experts decoder of DeepSeek-OCR, which activates about 570M parameters per token, with a FastMTP speculative decoding head that shares a single draft block recursively across K=3 prediction steps. Greedy verification makes decoding lossless. Post-training combines instruction alignment, robustness fine-tuning on difficult documents, and GRPO under dense verifiable rewards: deterministic formula, table, and structural checks that award partial credit. The training data mixes cleaned public corpora with targeted synthetic pages. At the default dynamic-resolution setting, Jina-OCR-v1 scores 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, and reaches the highest page throughput in our comparison at 2.57 pages per second. On a low-budget GPU such as the NVIDIA L4, FastMTP doubles decoding speed over greedy autoregressive decoding. The model is publicly available at https://huggingface.co/jinaai/jina-ocr-v1.
