Replacing Training with Memory: Listwise Selection for Text-to-SQL
Summary
MaP-SQL enhances Text-to-SQL listwise selection without fine-tuning by employing structured memory retrieval and permutation-based ranking aggregation, outperforming state-of-the-art methods with reduced token usage.
View Cached Full Text
Cached at: 09/04/26, 03:58 AM
Paper page - Replacing Training with Memory: Listwise Selection for Text-to-SQL
Source: https://huggingface.co/papers/2609.00834
Abstract
MaP-SQL improves listwise Text-to-SQL selection without fine-tuning by using structured memory retrieval and permutation-based ranking aggregation.
ModernText-to-SQLsystems often follow generate-execute-select pipelines, generating multiple candidate queries then selecting the best one.Listwise selection, by jointly comparing multiple candidates, has been widely adopted, but fine-tuning listwise selectors is costly. We thus propose a fine-tuning-free listwise selector. We replace two major fine-tuning objectives withinference-time strategies: (1) learning selection criteria as ordering and (2) mitigatingpositional bias. First, we build reusablestructured memoriesinstead of learning selection behavior as model parameters. Given a question, MaP-SQL retrieves memories distilled from training data that encode how natural language maps to schema elements, SQL operations, and expected outputs. These memories serve as explicit decision criteria for evaluating candidates in a listwise manner. Second, to mitigate ordering bias of listwise selectors, we aggregate rankings across multiple input permutations, with inference cost optimized by execution results and pointwise scoring. Our approach improves selection accuracy while maintaining efficiency and compatibility with existing large language models. AcrossText-to-SQLbenchmarks, it produces more stable selection without fine-tuning and fewer unnecessary comparisons than existing methods. On BIRD-dev, it outperforms the previous state-of-the-art selector-based method R^3-SQL by 2.02execution accuracypoints on average using the same candidate sets, with 2.92x fewer tokens.
View arXiv pageView PDFAdd to collection
Get this paper in your agent:
hf papers read 2609\.00834
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2609.00834 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2609.00834 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2609.00834 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
Memory Architectures for Multi-Turn Text-to-SQL: A Benchmark and Empirical Study
This paper introduces EnterpriseMem-Bench, a multi-turn Text-to-SQL benchmark, and evaluates five frontier models across memory architectures, finding that stateless models collapse by the third turn and that working memory yields the largest gains.
Learning to Retrieve: Dual-Level Long-Term Memory for Text-to-SQL Agents
This paper proposes MERIT, a dynamic multi-horizon memory retrieval framework for interactive text-to-SQL agents that uses episode-level and turn-level memory with learned retrieval policies optimized via reinforcement learning and a process reward model for dense rewards. Experiments on BIRD-Interact and Spider2-Snow show that MERIT outperforms static and single-horizon dynamic baselines in success rate while requiring fewer interaction turns.
SDAM: Structure-Difference-Aware Memory Evolution for Complex Text-to-SQL
This paper proposes SDAM, a memory-based framework for complex Text-to-SQL that uses structure-difference aware reasoning, contradiction-aware reflection, and schema-grounded memory evolution to improve SQL generation. Experiments show modest gains on BIRD-dev and Spider-test benchmarks.
R^3-SQL: Ranking Reward and Resampling for Text-to-SQL
# Paper page - R^3-SQL: Ranking Reward and Resampling for Text-to-SQL Source: [https://huggingface.co/papers/2604.25325](https://huggingface.co/papers/2604.25325) ## Abstract R$^3$\-SQL addresses inconsistencies in scoring functionally equivalent SQL queries and improves candidate recall through unified reward ranking and agentic resampling techniques\. Modern[Text\-to\-SQL](https://huggingface.co/papers?q=Text-to-SQL)systems generate multiple candidate[SQL queries](https://huggingface.co/papers
SERL-SQL: Selective Hindsight Distillation for Text-to-SQL Reinforcement Agentic Learning
SERL-SQL proposes a selective execution-grounded reinforcement learning framework for multi-turn Text-to-SQL agents, using teacher-student likelihood gaps to reweight GRPO advantages on SQL action tokens. It achieves strong results on BIRD and Spider benchmarks.