davanstrien/test-olmocr2
收藏Hugging Face2025-10-23 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/davanstrien/test-olmocr2
下载链接
链接失效反馈官方服务：
资源简介：
这是一个使用olmOCR-2-7B模型从[davanstrien/test-olmocr2](https://huggingface.co/datasets/davanstrien/test-olmocr2)数据集中的图像生成的markdown格式OCR结果的数据集。该数据集包含100个样本，处理时间为0小时3分32秒，处理日期为2025年10月23日。olmOCR-2-7B是一个基于Qwen2.5-VL-7B-Instruct的高质量文档OCR模型，经过olmOCR-mix-1025数据集的微调和GRPO强化学习优化。模型具有识别LaTeX公式、HTML表格、文档结构、图像描述、文档方向、自然阅读顺序等功能，准确率达到82.4 ± 1.1。
This dataset contains markdown-formatted OCR results from images in [davanstrien/test-olmocr2](https://huggingface.co/datasets/davanstrien/test-olmocr2) using the olmOCR-2-7B model. It includes 100 samples processed in 0h 3m 32s on October 23, 2025. olmOCR-2-7B is a high-quality document OCR model based on Qwen2.5-VL-7B-Instruct, fine-tuned on the olmOCR-mix-1025 dataset and optimized with GRPO reinforcement learning. The model features include LaTeX equation recognition, HTML table extraction, document structure preservation, figure descriptions, rotation detection, natural reading order handling, and high accuracy with a score of 82.4 ± 1.1 on olmOCR-Bench.
提供机构：
davanstrien
