hoadm/vibird

ViBIRD是BIRD文本到SQL基准的越南语翻译版本，包含问题和证据字段的越南语翻译。数据集分为训练集（9,428条）和开发集（1,534条），总计10,962条。每个条目包含唯一标识符、数据库标识符、原始英文问题和证据、SQL查询、SQL复杂度类、难度级别、越南语翻译问题和证据以及翻译方法。翻译方法包括人工翻译（1,415条）、GPT少量样本翻译（2,357条）和基于Qwen2.5-7B的双字段翻译器（5,656条）。数据集还提供了相关的模型、姊妹数据集和源代码链接。

ViBIRD is a Vietnamese translation of the BIRD Text-to-SQL benchmark, including Vietnamese translations of the question and evidence fields. The dataset is divided into a training set (9,428 items) and a development set (1,534 items), totaling 10,962 items. Each item includes a unique identifier, database identifier, original English question and evidence, SQL query, SQL complexity class, difficulty level, Vietnamese translated question and evidence, and translation method. Translation methods include human translation (1,415 items), GPT few-shot translation (2,357 items), and a fine-tuned Qwen2.5-7B dual-field translator (5,656 items). The dataset also provides links to related models, sister datasets, and source code.