hoadm/vispider

ViSpider是Spider文本到SQL基准（Yu等人，EMNLP 2018）的越南语翻译版本。该数据集用于表格问答和文本生成任务，特别是针对越南语的文本到SQL转换。数据集包含8,659个训练样本、1,034个开发样本和2,147个测试样本，总计11,840个样本。每个样本包括唯一标识符、数据库标识符、原始英文问题、黄金SQL查询、SQL复杂度类别、越南语问题翻译及翻译方法。翻译方法包括人工翻译（1,299个样本）、GPT少样本翻译（2,165个样本）和基于Qwen2.5-7B的微调翻译（5,195个样本）。

ViSpider is a Vietnamese translation of the Spider Text-to-SQL benchmark (Yu et al., EMNLP 2018). The dataset is designed for table-question-answering and text-generation tasks, specifically focusing on text-to-SQL conversion in Vietnamese. It includes 8,659 training samples, 1,034 development samples, and 2,147 test samples, totaling 11,840 samples. Each sample contains a unique identifier, database identifier, original English question, gold SQL query, SQL complexity class, Vietnamese question translation, and translation method. Translation methods include human translation (1,299 samples), GPT few-shot translation (2,165 samples), and fine-tuned Qwen2.5-7B translation (5,195 samples).