aisingapore/multiturn_chat-mtbench

SEA-MTBench数据集用于评估模型在多轮（两轮）对话中的表现，并确保其响应符合人类需求。该数据集基于MT-Bench，并由母语者手动翻译为印尼语（id）、爪哇语（jv）、巽他语（su）和越南语（vi）。泰语部分使用了来自ThaiLLM排行榜的MT-Bench Thai数据集。数据集包含英语、印尼语、爪哇语、巽他语、泰语和越南语的分割，每个分割都有详细的统计信息，包括示例数量和不同模型的token数量。

SEA-MTBench dataset is used to evaluate a models performance in multi-turn conversations, specifically how well the model meets human needs in two-turn dialogues. The dataset is based on MT-Bench and manually translated by native speakers into Indonesian, Javanese, Sundanese, and Vietnamese. The Thai portion uses the MT-Bench Thai dataset. The dataset contains multiple splits in different languages, each with specific numbers of examples and tokens. The main task of the dataset is text generation, suitable for evaluating chat or instruction-tuned large language models.