Progress-SQL: Improving Reinforcement Learning for Text-to-SQL via Progressive Rewards

2026-06-08 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, extended

Summary

Progress-SQL is a multi-turn reinforcement learning framework designed to enhance Text-to-SQL generation by large language models. It introduces an Oracle-guided Diagnostic Tree (ODT) that provides clause-level structural feedback for iterative SQL refinement. The framework utilizes a progressive reward system, combining ODT-based structural alignment with lexical alignment, a progression latency reward for earlier correctness, and an execution status reward for invalid SQL recovery. Experiments on BIRD, Spider, and Spider robustness variants demonstrate Progress-SQL improves 7B backbone models by an average of 8.5% in execution accuracy and 6.3% in test-suite accuracy on Spider Dev, consistently outperforming existing RL methods.

Key takeaway

For machine learning engineers developing Text-to-SQL solutions, adopting a multi-turn reinforcement learning approach with progressive rewards can significantly enhance model accuracy and robustness. You should consider integrating Oracle-guided Diagnostic Trees (ODT) to provide fine-grained, clause-level feedback during training, fostering iterative SQL refinement. This strategy improves first-attempt SQL generation and recovery from invalid queries, leading to more reliable systems.

Key insights

Multi-turn reinforcement learning with progressive rewards and ODT feedback significantly improves Text-to-SQL generation.

Principles

• Multi-turn RL with diagnostic feedback enhances SQL refinement.
• Progressive rewards capture trajectory-level improvement.
• Early correctness and execution recovery are crucial for robust RL.

Method

Progress-SQL uses multi-turn rollouts, generating SQLs iteratively. ODT provides clause-level structural feedback. Rewards combine progressive alignment, latency, execution status, and format.

In practice

• Implement ODT for fine-grained SQL error diagnosis.
• Design rewards for trajectory improvement, not just final state.
• Incorporate latency and executability for robust RL training.

Topics

• Text-to-SQL
• Reinforcement Learning
• Large Language Models
• SQL Generation
• Oracle-guided Diagnostic Tree
• Progressive Rewards

Code references

• YooYoo67/ProgressSQL
• RUCKBReasoning/OmniSQL

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• SIRIUS-SQL: Anchoring Multi-Candidate Text-to-SQL in Execution Feedback — Multi-candidate Text-to-SQL reliability improves by diversifying generation, targeting repairs, and hybrid selection.
• Progress-SQL: Improving Reinforcement Learning for Text-to-SQL via Progressive Rewards — Progress-SQL enhances Text-to-SQL reinforcement learning through progressive, multi-turn rewards and an Oracle-guided Diagnostic Tree for iterative SQL refinement.
• Text-to-SQL solution powered by Amazon Bedrock — Text-to-SQL solutions using Amazon Bedrock democratize data access by translating natural language questions into actionable insights.
• Residual Skill Optimization for Text-to-SQL Ensembles — DivSkill-SQL optimizes agent skills on residual failures to build complementary Text-to-SQL ensembles, improving Pass@K without model fine-tuning.
• A Reasoner for Real-World Event Detection: Scaling Reinforcement Learning via Adaptive Perplexity-Aware Sampling Strategy — APARL uses perplexity-aware reinforcement learning and dynamic curriculum to improve abnormal event detection and OOD generalization.
• ReTreVal: Reasoning Tree with Validation and Cross-Problem Memory for Large Language Models — Combining structured exploration, iterative refinement, explicit validation, and persistent memory significantly enhances LLM multi-step reasoning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.