ilsp/m-ArenaHard_greek

这是一个希腊语的m-ArenaHard数据集，是LMArena的arena-hard-auto-v0.1版本的翻译版。该数据集由Cohere的m-ArenaHard版本翻译而来，原始翻译使用Google Translate API v3完成。我们对数据集进行了进一步筛选，使用Claude Sonnet 3.5 v2对原始的Google Translate API v3提供的翻译进行了后编辑，因为注意到一些翻译（尤其是与编码相关的提示）不够准确。最终的数据集包含500个具有挑战性和多样性的提示，可以用来评估任何微调后的LLM模型的希腊语聊天能力。为了正确使用这个测试集，需要设置一个评判模型和一个基线模型。该数据集由ILSP/Athena RC策划，支持的语言为希腊语和英语，遵循Apache-2.0许可证。

This is the Greek version of the m-ArenaHard dataset, translated from LMArenas arena-hard-auto-v0.1. It originates from Coheres m-ArenaHard, which was originally translated using Google Translate API v3. We further curated the dataset by post-editing the original translations provided with Google Translate API v3 using Claude Sonnet 3.5 v2, as we noticed that some translations (especially those related to coding) were not accurate. The resulting dataset consists of 500 challenging and diverse prompts, which can be used to evaluate any fine-tuned LLM on its Greek chat capabilities. It is necessary to set a judge model and a baseline model to properly use this test set. The dataset was curated by ILSP/Athena RC, supports Greek and English languages, and is licensed under Apache-2.0.