baconnier/deepsynth-fr

这是一个大规模的法语新闻摘要数据集，来源于主要的法国报纸，用于训练多语言的DeepSeek-OCR模型，并且能够正确处理Unicode字符和重音符号。数据集是DeepSynth项目的一部分，该项目使用视觉文本编码技术进行多语言摘要，将文本文档转换为图像并通过视觉编码器处理，实现了20倍的token压缩比例，同时保持了文档的布局和结构。

This is a large-scale French news summarization dataset from major French newspapers, designed for training multilingual DeepSeek-OCR models with proper Unicode/diacritics handling. It is part of the DeepSynth project, which uses visual text encoding for multilingual summarization, converting text documents into images and processing them through a visual encoder to achieve a 20x token compression ratio while preserving document layout and structure.