Memo-SQL: Structured Decomposition and Experience-Driven Self-Correction for Training-Free NL2SQL

TL;DR

Memo-SQL introduces a training-free approach for NL2SQL that leverages structured decomposition and experience-driven self-correction, significantly improving accuracy and efficiency without fine-tuning.

Contribution

It proposes a novel, training-free framework combining structured question decomposition and dynamic memory for self-correction in NL2SQL tasks.

Findings

Achieves 68.5% execution accuracy on BIRD dataset.

Outperforms previous open, zero-fine-tuning methods.

Uses over 10 times fewer resources than prior approaches.

Abstract

Existing NL2SQL systems face two critical limitations: (1) they rely on in-context learning with only correct examples, overlooking the rich signal in historical error-fix pairs that could guide more robust self-correction; and (2) test-time scaling approaches often decompose questions arbitrarily, producing near-identical SQL candidates across runs and diminishing ensemble gains. Moreover, these methods suffer from a stark accuracy-efficiency trade-off: high performance demands excessive computation, while fast variants compromise quality. We present Memo-SQL, a training-free framework that addresses these issues through two simple ideas: structured decomposition and experience-aware self-correction. Instead of leaving decomposition to chance, we apply three clear strategies, entity-wise, hierarchical, and atomic sequential, to encourage diverse reasoning. For correction, we build a dynamic memory of both successful queries and historical error-fix pairs, and use retrieval-augmented prompting to bring relevant examples into context at inference time, no fine-tuning or external APIs required. On BIRD, Memo-SQL achieves 68.5% execution accuracy, setting a new state of the art among open, zero-fine-tuning methods, while using over 10 times fewer resources than prior TTS approaches.