OXR:解析失败纳入训练,开源文档解析速度超GLM-OCR三倍
OXR makes parsing failures part of the training process.
开源文档解析领域的新标杆,不仅刷新了多项基准测试纪录,更通过 RSI 机制实现了性能与速度的双重突破,做 OCR 或文档处理的同学建议直接参考其架构思路。
OXR makes parsing failures part of the training process.
OXR 将解析失败纳入训练流程。
During post-training (OXR is the first document parser to use RSI here), a VLM evaluates its outputs, rollout disagreements flag uncertainty, and a separate clustering model groups the causes;
在训练后阶段(OXR 是首个在此处使用 RSI 的文档解析器),VLM 会评估其输出,rollout 分歧标记不确定性,而独立的聚类模型则对原因进行分组;
those findings then guide targeted data generation for the next round, and the 1.3B model that comes out of it sits first on olmOCR-Bench, OmniDocBench and DocParsingBench while running 3.05× GLM-OCR's throughput on a single RTX PRO 6000, the fastest open-source result in the speed comparison.
这些发现随后指导下一轮的目标数据生成,由此产生的 1.3B 模型在 olmOCR-Bench、OmniDocBench 和 DocParsingBench 上均位列第一,同时在单张 RTX PRO 6000 显卡上的吞吐量达到 GLM-OCR 的 3.05 倍,成为速度对比中最快的开源结果。
Worth studying (Github link in comment)
值得研究(评论中有 Github 链接)
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力