Jina-OCR-v1用高效解码和验证奖励，让低配显卡也能快速精准解析文档。
Jina-OCR-v1: Efficient Document Parsing with Speculative Decoding and Dense Verifiable Rewards
- 采用压缩视觉编码器与3B专家混合解码器，每词激活约5.7亿参数
- 动态分辨率下在OmniDocBench达91.14分，每秒处理2.57页
- 适合资源有限但需高精度文档解析的场景
我们提出Jina-OCR-v1，一个面向低成本GPU的端到端文档解析模型。它结合DeepSeek-OCR的压缩视觉编码器与3B专家混合解码器（每词激活约5.7亿参数），并引入FastMTP推测解码头，在K=3个预测步骤中复用单个草稿块，贪婪验证确保解码无损。后训练融合指令对齐、难文档鲁棒性微调及基于密集可验证奖励的GRPO：通过确定性公式、表格与结构检查给予部分得分。训练数据混合清洗后的公开语料与定向合成页面。在默认动态分辨率设置下，Jina-OCR-v1在OmniDocBench v1.6上得分为91.14，在olmOCR-Bench上为83.4，页处理吞吐量达2.57页/秒，领先于对比模型。在如NVIDIA L4等低预算显卡上，FastMTP使解码速度较贪心自回归解码翻倍。模型已开源：https://huggingface.co/jinaai/jina-ocr-v1。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。