PARROT
收藏资源简介:
PARROT是一个用于评估跨系统SQL翻译的实用且真实的基准数据集,它由来自38个开源基准和真实世界商业服务的598个翻译对组成,旨在挑战特定系统的SQL理解。它还提供了多个基准变体,包括PARROT-Diverse(包含28,003个翻译,用于广泛的语法测试)和PARROT-Simple(包含5,306个代表性样本,用于集中的压力测试),涵盖了22个生产级数据库系统。该数据集旨在促进未来研究,通过公开排行榜和源代码发布,供研究人员和实践者根据自己的需要定制评估。
PARROT is a practical and authentic benchmark dataset for evaluating cross-system SQL translation. It comprises 598 translation pairs sourced from 38 open-source benchmarks and real-world commercial services, aiming to challenge the SQL comprehension capabilities of specific database systems. It also offers multiple benchmark variants, including PARROT-Diverse (with 28,003 translation pairs for comprehensive syntax testing) and PARROT-Simple (with 5,306 representative samples for focused stress testing), covering 22 production-grade database systems. This dataset is intended to facilitate future research, with public leaderboards and source code released to allow researchers and practitioners to customize evaluations according to their respective needs.
PARROT 数据集概述
数据集简介
PARROT(Practical And Realistic Benchmark for Cross-System SQL Translation)是一个专为跨系统SQL翻译任务设计的基准数据集,用于支持将一个数据库系统编写的查询转换为功能等效的另一种数据库系统查询。
数据集规模
- 主要数据集包含598个翻译对
- 来源于38个开源基准和真实业务服务
- 专门设计用于挑战系统特定的SQL理解能力
扩展版本
- PARROT-1.0版本包含28,003个翻译对
- 同样来源于38个开源基准
- 用于广泛的语法测试
性能表现
- 不同LLM在SQL到SQL翻译任务中的平均准确率低于38.53%
- 人类专家(翻译工具+数据库管理员)的准确率超过90.00%
评估指标
方言兼容性(AccEX)
- GPT-4o:53.32%
- DeepSeek-V3 671B:50.64%
- Claude 3.7 Sonnet:48.09%
结果一致性(AccRES)
- o3-mini:54.23%
- o1-preview:48.69%
- DeepSeek-R1 671B:40.52%
引用文献
bibtex @inproceedings{zhou2025parrot, author = {Wei Zhou and Guoliang Li and Haoyu Wang and Yuxing Han and Xufei Wu and Fan Wu and Xuanhe Zhou}, title = {PARROT: A Benchmark for Evaluating LLMs in Cross-System SQL Translation}, booktitle = {NeurIPS}, year = {2025} }
资源获取
数据集及相关使用说明已在GitHub仓库公开发布。




