UnifiedCrawl-Amharic

该数据集是从Common Crawl语料库中提取的单语种数据集，专注于阿姆哈拉语。它采用了一种高效筛选和提取文本数据的方法。该数据集特别为资源较少的语言设计，尤其是阿姆哈拉语，并在语言建模和少量样本提示得分方面展示了改进。其规模超过了之前为资源较少语言提供的任何资源，其任务是微调多语言大型语言模型。

This is a monolingual dataset focused on Amharic, extracted from the Common Crawl corpus. It employs an efficient methodology for screening and extracting textual data. Developed specifically for low-resource languages—with Amharic as its primary target—this dataset has shown improved performance in language modeling and few-shot prompt scoring. It boasts a larger scale than any previously released resource for low-resource languages, and is intended for fine-tuning multilingual large language models.