laulauland/overpass-rl-train

--- license: cc-by-4.0 language: en tags: - overpass-ql - openstreetmap - synthetic - text-to-query --- # Overpass RL Training Prompts Synthetic natural language prompts for training an Overpass QL generation model via GRPO. ## Sources - **Synthetic prompts**: Generated by Claude Haiku 4.5 from OSM taginfo corpus tag combinations with varied spatial patterns, query types, and phrasing styles. - **OverpassNL train split**: Natural language inputs from the OverpassNL dataset (without gold queries), mixed in for ecological validity. See [OverpassNL](https://github.com/raphael-sch/OverpassNL): > Staniek, M., Fink, N., Mahdavi, S. A., & Riezler, S. (2024). > Text-to-OverpassQL: A Natural Language Interface for Complex Geodata Querying of OpenStreetMap. > TACL, 12, 1375-1390. [arXiv:2308.16060](https://arxiv.org/abs/2308.16060) - **Tag corpus**: Built from [taginfo.openstreetmap.org](https://taginfo.openstreetmap.org) database dumps. OSM data is licensed under [ODbL](https://opendatacommons.org/licenses/odbl/). ## Schema - `prompt`: Natural language request (no system prompt, no gold query)