ReToolSQL: Agentic Reinforcement Learning for Robust Text-to-SQL

• URL: http://arxiv.org/abs/2608.27796v1
• Date: Fri, 28 Aug 2026 00:23:31 GMT
• Title: ReToolSQL: Agentic Reinforcement Learning for Robust Text-to-SQL
• Abstract summary: We present ReTool, a two-stage training framework for text-to-teacher.<n>We show that a properly designed SFT$to$RFT pipeline over tool-use trajectories is a practical path toward robust enterprisegrade text-to-teacher.
• Score: 3.318897003817309
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Recent work has shown that reinforcement learning from execution feedback can substantially improve text-to-SQL performance, often enabling smaller models to match or exceed much larger systems. However, most existing approaches treat SQL generation as a single-turn task, limiting the model's ability to recover from errors through iterative refinement. We present ReToolSQL, a two-stage training framework for text-to-SQL that combines (i) a supervised warm-start on rejection-sampled reasoning traces with (ii) agentic reinforcement fine-tuning (RFT) over multi-turn tool-use trajectories. The key insight is that the two stages act on complementary axes, the supervised fine-tuning (SFT) on verified privileged-teacher traces expands the set of solvable questions (raising pass@k coverage on the hardest cases), while RFT converts that expanded capability into higher single-pass accuracy by teaching the model when to verify, what evidence to retrieve, and how to repair faulty SQL from execution feedback. Applied to Gemma 4 instruction-tuned (31B), RFT alone achieves 73.66% execution accuracy (EX) on the BIRD-SQL development benchmark (74.12% EX with self-consistency). Initializing RFT from the SFT checkpoint (SFT$\to$RFT) yields our strongest model at 74.32% EX single-pass and 74.77% EX with self-consistency. At the time of writing, this ranked first on the BIRD single-model development-set leaderboard. The approach uses composite rewards anchored on execution correctness, requires no human annotation beyond the benchmark itself, and operates within a single dense 31B model, showing that a properly designed SFT$\to$RFT pipeline over tool-use trajectories is a practical path toward robust enterprise-grade text-to-SQL.

Related papers

• Spider 2.0-AIFunc: Extending Real-World Text-to-SQL to AI-Native SQL Workflows [54.49251107006261]
  We introduce Spider 2.0-AIFunc, a benchmark of 465 verified instances across 125 real-world databases covering six types of AI functions on the Snowflake platform.<n>We find that the strongest proprietary models reach 67-70% execution accuracy while the best open-source model achieves 58.1%.
  arXiv  Detail & Related papers  (2026-07-07T12:54:08Z)
• ReViSQL: Achieving Human-Level Text-to-SQL [8.94428202485629]
  We introduce ReVi, a streamlined framework to improve human-level accuracy on BIRD benchmark data.<n>Instead of complex AI agents, ReVi leverages learning with verifiable rewards (RLVR) on BIRD-, a dataset we curated.<n>We identify and corrected data errors in 61.1% of a subset of BIRD Train and show that improving data quality alone boosts the single-generation accuracy by 8.2-13.9%.
  arXiv  Detail & Related papers  (2026-03-20T14:49:27Z)
• SQL-Trail: Multi-Turn Reinforcement Learning with Interleaved Feedback for Text-to-SQL [20.49395306069103]
  We introduce a multi-turn reinforcement learning (RL) agentic framework for Text-to-one generation.<n>Rather than producing a query in one shot, SQL-Trail interacts with the database environment and uses execution feedback to iteratively refine its predictions.<n>Our approach centers on two key ideas: (i) an adaptive turn-budget allocation mechanism that scales the agent's interaction depth to match question difficulty, and (ii) a composite reward panel that jointly incentivizessql correctness and efficient exploration.
  arXiv  Detail & Related papers  (2026-01-25T05:16:52Z)
• Text-to-SQL as Dual-State Reasoning: Integrating Adaptive Context and Progressive Generation [54.53145282349042]
  We introduce DSR-sourced, a textbfDual-textbfS textbfReasoning framework that models Text-to-context as an interaction between an adaptive context state and a progressive generation state.<n>Without any post-training or in-context examples, DSR-sourced achieves competitive performance, reaching 35.28% execution accuracy on Spider 2.0-Snow and 68.32% on BIRD development set.
  arXiv  Detail & Related papers  (2025-11-26T13:52:50Z)
• MTIR-SQL: Multi-turn Tool-Integrated Reasoning Reinforcement Learning for Text-to-SQL [46.37961458768655]
  Large language models (LLMs) are increasingly used in Text-to-aware tasks.<n>Existing methods rely on static execution feedback, which restricts real-time error correction.<n>We propose MTIR-IDER, an innovative Multi-turn Tool-Integrated Reasoning reinforcement learning framework.
  arXiv  Detail & Related papers  (2025-10-29T13:34:27Z)
• CogniSQL-R1-Zero: Lightweight Reinforced Reasoning for Efficient SQL Generation [1.169202600932732]
  We introduce Cogni-R1-Zero, a reinforcement learning (RL) framework and model.<n>We use a lightweight reward signal based on execution correctness and format-tag compliance.<n>Our method achieves state-of-the-art execution accuracy on Text2 benchmark.<n>To support further research in efficient and interpretable Text-to-code modeling, we release two curated datasets.
  arXiv  Detail & Related papers  (2025-07-08T14:17:07Z)
• RAISE: Reasoning Agent for Interactive SQL Exploration [47.77323087050061]
  We propose a novel framework that unifies schema linking, query generation, and iterative refinement within a single, end-to-end component.<n>Our method emulates how humans answer questions when working with unfamiliar databases.
  arXiv  Detail & Related papers  (2025-06-02T03:07:08Z)
• Arctic-Text2SQL-R1: Simple Rewards, Strong Reasoning in Text-to-SQL [35.21185734929167]
  We present Arctic-Text2-R1, a reinforcement learning (RL) framework and model family designed to generate accurate, executablesql.<n>Our approach avoids curated intermediate supervision and complex reward shaping, promoting stable training and alignment with the end task.<n> Notably, our 7B model outperforms prior 70B-class systems, highlighting the framework's scalability and efficiency.
  arXiv  Detail & Related papers  (2025-05-22T23:33:47Z)
• ReEx-SQL: Reasoning with Execution-Aware Reinforcement Learning for Text-to-SQL [21.29343406227956]
  Execution feedback is essential for guiding large language models to reason accurately and generate reliable queries.<n>Existing methods treat execution feedback solely as a post-hoc signal for correction or selection, failing to integrate it into the generation process.<n>We propose ReEx-Reasoning, a framework for Text-to-Reason that enables models to interact with the database during decoding and dynamically adjust their reasoning based on execution feedback.
  arXiv  Detail & Related papers  (2025-05-19T06:46:47Z)
• ExCoT: Optimizing Reasoning for Text-to-SQL with Execution Feedback [49.21833666405111]
  Large language models (LLMs) excel in many reasoning tasks, but their ability to leverage Chain-of-Thought (CoT) reasoning remains underexplored.<n>We propose ExCoT, a novel framework that iteratively optimize open-source LLMs by combining CoT reasoning with off-policy and on-policy DPO.
  arXiv  Detail & Related papers  (2025-03-25T18:17:36Z)
• RSL-SQL: Robust Schema Linking in Text-to-SQL Generation [51.00761167842468]
  We propose a novel framework called RSL- that combines bidirectional schema linking, contextual information augmentation, binary selection strategy, and multi-turn self-correction. benchmarks demonstrate that our approach achieves SOTA execution accuracy among open-source solutions, with 67.2% on BIRD and 87.9% on GPT-4ocorrection. Our approach outperforms a series of GPT-4 based Text-to-Seek systems when adopting DeepSeek (much cheaper) with same intact prompts.
  arXiv  Detail & Related papers  (2024-10-31T16:22:26Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.