Research map / Large language models / Language, safety and interpretability
RetrySQL: Text-to-SQL Training with Retry Data for Self-Correcting Query Generation
Alicja Rączkowska, Riccardo Belluzzo, Piotr Zieliński, Joanna Baran, Paweł Olszewski
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add retry data (FS, p_retry 0.2) to continued pre-training instead of error-free reasoning (ablation) | OpenCoder 1.5B · BIRD dev (perfect schema linking) | EX overall: 54.71 -> 58.70 | Improved |
| add retry data (FS 0.2) instead of error-free reasoning (ablation) | OpenCoder 1.5B · SPIDER dev (perfect schema linking) | EX overall: 74.25 -> 77.35 | Improved |
| add retry data (FS 0.3) instead of error-free reasoning (ablation) | OpenCoder 1.5B · BIRD dev challenging | EX challenging: 39.45 -> 43.31 | Improved |
| add retry data (FS 0.3) instead of error-free reasoning (ablation) | Qwen2.5-Coder 1.5B · BIRD dev (perfect schema linking) | EX overall: 55.96 -> 56.36 | No clear effect |
| add retry data (FS 0.2) instead of error-free reasoning (ablation) | Qwen2.5-Coder 1.5B · BIRD dev (perfect schema linking) | EX overall: 55.96 -> 55.65 | Hurt |
| add retry data (FS 0.2) instead of error-free reasoning (ablation) | Qwen2.5-Coder 1.5B · SPIDER dev (perfect schema linking) | EX overall: 73.25 -> 77.18 | Improved |
| add retry data (FS 0.3) instead of error-free reasoning (ablation) | Qwen2.5-Coder 1.5B · BIRD dev challenging | EX challenging: 30.21 -> 39.10 | Improved |
| add retry data (FS 0.3) on SPIDER hard questions (ablation) | OpenCoder 1.5B · SPIDER dev hard | EX hard: 69.88 -> 62.76 | Hurt |
| continued pre-training with reasoning-augmented text-to-SQL data instead of zero-shot | OpenCoder 1.5B · BIRD dev (perfect schema linking) | EX overall: 38.44 -> 54.71 | Improved |
| use FS (forward single) corruption instead of other retry variants (ablation) | RetrySQL · BIRD / SPIDER dev | EX overall | Improved |
| use p_retry of 0.2 or 0.3 (ablation) | RetrySQL · BIRD / SPIDER dev | EX overall | Improved |
Closest papers by meaning
- SQLens: An End-to-End Framework for Error Detection and Correction in Text-to-SQL (NeurIPS 2025 · similarity 0.89)
- ROUTE: Robust Multitask Tuning and Collaboration for Text-to-SQL (ICLR 2025 · similarity 0.89)
- Alpha-SQL: Zero-Shot Text-to-SQL using Monte Carlo Tree Search (ICML 2025 · similarity 0.89)
- CogSQL: A Cognitive Framework for Enhancing Large Language Models in Text-to-SQL Translation (AAAI 2025 · similarity 0.88)
- Structure-Guided Large Language Models for Text-to-SQL Generation (ICML 2025 · similarity 0.88)
- SQL-R1: Training Natural Language to SQL Reasoning Model By Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
