--- title: TERN Benchmark sdk: static app_file: index.html fullWidth: true short_description: Time-series text-to-SQL benchmark on Apache IoTDB --- # TERN TERN is a time-series text-to-SQL benchmark generated by TSSDA and built on Apache IoTDB. This Space hosts the project homepage, EX leaderboard, dataset examples, and email submission form for organizer review. - [Paper](https://arxiv.org/abs/2611.02132) - [Code](https://github.com/chris265100/TSSDA) The download package includes 10,731 train NL–SQL pairs, 1,326 dev NL–SQL pairs, 1,125 test questions, all 32 databases, schemas, Apache IoTDB 2.0.5 and one-command deployment instructions. Test gold SQL is withheld; official evaluation requires submitting a request to the benchmark maintainers. Preserve the integer question IDs from test.json when preparing predictions. Download `downloads/TERN.zip` from the homepage. `downloads/TERN.zip.sha256` provides its SHA-256 checksum. The website does not execute SQL or provide online EX evaluation. Email submissions are forwarded through FormSubmit for organizer review. Local file validation only checks the prediction format. Download request statistics are supplied by an API with persistent storage. The count covers requests through the tracked website button, excludes HEAD and prefetch requests, and does not measure completed file transfers or unique people. Historical downloads and direct archive links are not included. This is a buildless HTML/CSS/JavaScript Space; no Python server, database server, Docker image, GPU, or package installation is needed to serve the website.