Geraldine/mlintern-gemma4-unimarc-training

Gemma-4 UNIMARC训练数据集旨在微调google/gemma-4-E2B-it模型，从书籍和论文封面图像中提取书目元数据并生成有效的UNIMARC记录。该数据集结合了来自两个来源的真实书目数据和合成增强数据。数据集包含详细的来源信息、格式、覆盖的UNIMARC字段、训练脚本、依赖项、数据来源、许可和参考文献。该数据集专门用于训练视觉语言模型（VLM）以实现自动化的UNIMARC编目。

The Gemma-4 UNIMARC Training Dataset is designed for fine-tuning the google/gemma-4-E2B-it model to extract bibliographic metadata from book and thesis cover images and generate valid UNIMARC records. The dataset combines real bibliographic data from two sources with synthetic augmentation. It includes detailed information about the datasets sources, format, UNIMARC fields covered, training script, dependencies, data sources, licensing, and references. The dataset is specifically tailored for training Vision-Language Models (VLMs) for automated UNIMARC cataloguing.