![[2602.15564] Beyond Static Pipelines: Learning Dynamic Workflows for Text-to-SQL](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.15564] Beyond Static Pipelines: Learning Dynamic Workflows for Text-to-SQL
Summary
The paper presents a novel approach to Text-to-SQL systems by introducing dynamic workflows that adapt during inference, enhancing performance over static methods.
Why It Matters
This research addresses the limitations of static workflows in Text-to-SQL applications, which often struggle with real-world complexities. By proposing a reinforcement learning framework (SquRL) for dynamic workflow construction, the authors aim to improve adaptability and scalability, making Text-to-SQL systems more effective in diverse scenarios.
Key Takeaways
- Dynamic workflows outperform static methods in Text-to-SQL tasks.
- The proposed SquRL framework enhances reasoning capabilities of LLMs.
- Innovative training mechanisms improve exploration and efficiency.
- Performance gains are particularly notable in complex queries.
- The research provides a pathway for better adaptability in AI systems.
Computer Science > Computation and Language arXiv:2602.15564 (cs) [Submitted on 17 Feb 2026] Title:Beyond Static Pipelines: Learning Dynamic Workflows for Text-to-SQL Authors:Yihan Wang, Peiyu Liu, Runyu Chen, Wei Xu View a PDF of the paper titled Beyond Static Pipelines: Learning Dynamic Workflows for Text-to-SQL, by Yihan Wang and 3 other authors View PDF HTML (experimental) Abstract:Text-to-SQL has recently achieved impressive progress, yet remains difficult to apply effectively in real-world scenarios. This gap stems from the reliance on single static workflows, fundamentally limiting scalability to out-of-distribution and long-tail scenarios. Instead of requiring users to select suitable methods through extensive experimentation, we attempt to enable systems to adaptively construct workflows at inference time. Through theoretical and empirical analysis, we demonstrate that optimal dynamic policies consistently outperform the best static workflow, with performance gains fundamentally driven by heterogeneity across candidate workflows. Motivated by this, we propose SquRL, a reinforcement learning framework that enhances LLMs' reasoning capability in adaptive workflow construction. We design a rule-based reward function and introduce two effective training mechanisms: dynamic actor masking to encourage broader exploration, and pseudo rewards to improve training efficiency. Experiments on widely-used Text-to-SQL benchmarks demonstrate that dynamic workflow construction co...
