Instruction tuning for Text-to-SQL works well with less data
LIMIT: Less Is More for Instruction Tuning in Text-to-SQL
Artificial Intelligence
Summary
Converting questions into database queries usually requires training large language models with lots of examples. This paper finds that carefully choosing a small set of examples is more effective than using a large amount of data. The authors developed a method called LIMIT that picks the most useful samples based on difficulty, consistency, diversity, and quality. With only a few hundred examples, they matched or beat performance of systems trained on much more data.
What this means in practice
- •For natural language interface developers: Build efficient text-to-SQL systems by training with a small, carefully chosen dataset instead of very large corpora.
- •For enterprise database teams: Improve automation in querying internal databases by fine-tuning language models using fewer, high-quality instruction examples.
Authors
Haoyuan Ma, Hengwei Liu, Linjuan Wu, Yongliang Shen, Weiming Lu
Abstract
Large language models have achieved remarkable progress on Text-to-SQL through reasoning-enhanced fine-tuning, yet existing approaches predominantly rely on massive instruction corpora under the assumption that scale drives performance. We challenge this paradigm by investigating a fundamental question: what is the minimal data requirement for effective Text-to-SQL instruction tuning? We propose LIMIT(Less Is More for Instruction Tuning in Text-to-SQL), a data-centric framework that demonstrates strong database reasoning can emerge from an extremely compact training set when examples are strategically selected. LIMIT operates through four stages: difficulty-aware filtering that identifies samples within the model's learning frontier, chain-of-thought synthesis with consistency-based selection, multi-dimensional quality scoring via LLM-as-judge, and genetic algorithm optimization that jointly maximizes schema coverage and sample quality. On the BIRD and Spider benchmark, LIMIT selects only 796 and 863 samples while achieving 100% table coverage, enabling Qwen3-8B to reach 69.1% and 88.9% execution accuracy.This result surpasses methods trained on 20 times more data and establishes a new state-of-the-art among open-source approaches. Our findings suggest that careful data curation, rather than scale, is the key to efficient Text-to-SQL learning.