R$^3$-SQL: Ranking Reward and Resampling for Text-to-SQL

• URL: http://arxiv.org/abs/2604.25325v1
• Date: Tue, 28 Apr 2026 07:40:50 GMT
• Title: R$^3$-SQL: Ranking Reward and Resampling for Text-to-SQL
• Abstract summary: R$3$- achieves 75.03 execution accuracy on BIRD-dev, a new state of the art among methods using models with disclosed sizes, with consistent gains across five benchmarks.
• Score: 55.967050404665606
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Modern Text-to-SQL systems generate multiple candidate SQL queries and rank them to judge a final prediction. However, existing methods face two limitations. First, they often score functionally equivalent SQL queries inconsistently despite identical execution results. Second, ranking cannot recover when the correct SQL is absent from the candidate pool. We propose R$^3$-SQL, a Text-to-SQL framework that addresses both issues through unified reward for ranking and resampling. R$^3$-SQL first groups candidates by execution result and ranks groups for consistency. To score each group, it combines a pairwise preference across groups with a pointwise utility from the best group rank and size, capturing relative preference, consistency, and candidate quality. To improve candidate recall, R$^3$-SQL introduces agentic resampling, which judges the generated candidate pool and selectively resamples when the correct SQL is likely absent. R$^3$-SQL achieves 75.03 execution accuracy on BIRD-dev, a new state of the art among methods using models with disclosed sizes, with consistent gains across five benchmarks.

Related papers

• Replacing Training with Memory: Listwise Selection for Text-to-SQL [34.25214572225323]
  We build reusable structured memories instead of learning selection behavior as model parameters.<n>We aggregate rankings across multiple input permutations, with inference cost optimized by execution results and pointwise scoring.<n>Across Text-to- benchmarks, it produces more stable selection without fine-tuning and fewer unnecessary comparisons than existing methods.
  arXiv  Detail & Related papers  (2026-09-01T07:35:00Z)
• SIRIUS-SQL: Anchoring Multi-Candidate Text-to-SQL in Execution Feedback [26.80964126227997]
  Text-to-on complex schemas is unreliable on a single pass, so recent systems generate multiplesql candidates and let voting filter out errors.<n>Voting alone is not enough, because the multi-candidate recipe has three coupled weaknesses.<n>We present SIRIUS--32B, which addresses all three weaknesses.
  arXiv  Detail & Related papers  (2026-05-31T13:59:09Z)
• Residual Skill Optimization for Text-to-SQL Ensembles [37.13373780149542]
  We present DivSkill, a residual skill optimization framework that builds complementary agentic Text-to- ensembles without model fine-tuning.<n>Each new skill is optimized on examples the current skill ensemble fails on, provably targeting its marginal contribution to Pass@K.<n>On Spider2-Lite, DivSkill improves selected accuracy by up to +11.1 points on Snowflake and +8.3 on BigQuery over the strongest ensemble baseline.<n>Skills optimized a single dialect transfer without retraining across dialects (Snowflake, BigQuery, pts) and to a different task formulation, such as BIRD-Critic (+2.6)
  arXiv  Detail & Related papers  (2026-05-20T22:36:11Z)
• JudgeSQL: Reasoning over SQL Candidates with Weighted Consensus Tournament [26.476304887882133]
  Judge is a principled framework that redefines candidate selection through structured reasoning and weighted consensus tournament mechanism.<n>We show that Judge exhibits superior judgment capabilities and good cross-scale generalization and to generator capacity.
  arXiv  Detail & Related papers  (2025-10-17T11:46:38Z)
• PARROT: A Benchmark for Evaluating LLMs in Cross-System SQL Translation [21.0303026118673]
  We introduce PARROT, a practical and realistic benchmak for CrOss-System SQL Translation.<n> PARROT comprises 598 translation pairs from 38 open-source benchmarks and real-world business services.<n>We also provide multiple benchmark variants, including PARROT-Diverse with 28,003 translations and PARROT-Simple with 5,306 representative samples.
  arXiv  Detail & Related papers  (2025-09-27T14:41:13Z)
• XiYan-SQL: A Novel Multi-Generator Framework For Text-to-SQL [48.45491386478092]
  We present XiYan-, an innovative framework effectively generating and utilizing multiplesql candidates.<n>Overall, XiYan- achieves a new SOTA performance of 75.63% on the notable BIRD benchmark.<n>It also attains SOTA performance on the Spider test set with an accuracy of 89.65%.
  arXiv  Detail & Related papers  (2025-07-07T06:50:46Z)
• SQLFixAgent: Towards Semantic-Accurate Text-to-SQL Parsing via Consistency-Enhanced Multi-Agent Collaboration [26.193588535592767]
  We introduce a new consistency-enhanced multi-agent collaborative framework designed for detecting and repairing erroneous SQL.<n>We evaluate our proposed framework on five Text-to-text benchmarks.<n>Our method consistently enhances the performance of the baseline model.<n>Our framework has a higher token efficiency compared to other advanced methods, making it more competitive.
  arXiv  Detail & Related papers  (2024-06-19T09:57:19Z)
• SQLPrompt: In-Context Text-to-SQL with Minimal Labeled Data [54.69489315952524]
  "Prompt" is designed to improve the few-shot prompting capabilities of Text-to-LLMs. "Prompt" outperforms previous approaches for in-context learning with few labeled data by a large margin. We show that emphPrompt outperforms previous approaches for in-context learning with few labeled data by a large margin.
  arXiv  Detail & Related papers  (2023-11-06T05:24:06Z)
• Benchmarking and Improving Text-to-SQL Generation under Ambiguity [25.283118418288293]
  We develop a novel benchmark called AmbiQT where each text is interpretable as two plausible SQLs due to lexical and/or structural ambiguity. We propose LogicalBeam, a new decoding algorithm that navigates thesql logic space using a blend of plan-based template generation and constrained infilling.
  arXiv  Detail & Related papers  (2023-10-20T17:00:53Z)
• UNITE: A Unified Benchmark for Text-to-SQL Evaluation [72.72040379293718]
  We introduce a UNIfied benchmark for Text-to-domain systems. It is composed of publicly available text-to-domain datasets and 29K databases. Compared to the widely used Spider benchmark, we introduce a threefold increase in SQL patterns.
  arXiv  Detail & Related papers  (2023-05-25T17:19:52Z)
• Weakly Supervised Text-to-SQL Parsing through Question Decomposition [53.22128541030441]
  We take advantage of the recently proposed question meaning representation called QDMR. Given questions, their QDMR structures (annotated by non-experts or automatically predicted) and the answers, we are able to automatically synthesizesql queries. Our results show that the weakly supervised models perform competitively with those trained on NL- benchmark data.
  arXiv  Detail & Related papers  (2021-12-12T20:02:42Z)
• How to Query An Oracle? Efficient Strategies to Label Data [59.89900843097016]
  We consider the basic problem of querying an expert oracle for labeling a dataset in machine learning. We present a randomized batch algorithm that operates on a round-by-round basis to label the samples and achieves a query rate of $O(fracNk2)$. In addition, we present an adaptive greedy query scheme, which achieves an average rate of $approx 0.2N$ queries per sample with triplet queries.
  arXiv  Detail & Related papers  (2021-10-05T20:15:35Z)
• Bertrand-DR: Improving Text-to-SQL using a Discriminative Re-ranker [1.049360126069332]
  We propose a novel discnative re-ranker to improve the performance of generative text-to-rimi models. We analyze relative strengths of the text-to-rimi and re-ranker models for optimal performance. We demonstrate the effectiveness of the re-ranker by applying it to two state-of-the-art text-to-rimi models.
  arXiv  Detail & Related papers  (2020-02-03T04:52:47Z)

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.