FALCON: A Model and Dataset Agnostic Framework for Synthetic Data Generation for NL2SQL Pairs
FALCON generates realistic, ambiguity-aware NL2SQL training pairs with compact open models, matching or exceeding challenging benchmark complexity locally.
FALCON combines reserved-word SQL seeding, persona-based prompting, and alignment-based filtering to generate structurally complex, ambiguity-aware NL-to-SQL data at low cost using compact open models. Human evaluation confirms consistent high quality across model sizes, and the generated data exceeds existing benchmarks in both SQL complexity and natural-language richness. Models trained on FALCON data increasingly outperform baseline-trained models as query complexity rises, and mixing in a small proportion of benchmark data recovers performance on simpler queries. The model- and database-agnostic design lets organizations generate high-complexity training data locally without external APIs.
- Ambiguity-aware synthetic NL2SQL generation using compact open models
- Alignment-based filtering distinguishes incorrect from complex-but-valid queries
- FALCON-trained models outperform baselines on complex queries
- Local generation works without external APIs
- Organizations
- FALCON
From the original article
1 min read at arxiv.orgRelational databases are among the most widely deployed forms of structured knowledge, and natural language access to them requires grounding language onto schema entities and relations while handling the ambiguity inherent in how people phrase requests.
By Darian Lee · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.