TrustSQL: Benchmarking Text-to-SQL Reliability with Penalty-Based Scoring

TL;DR

TrustSQL introduces a benchmark and scoring system to evaluate and improve the reliability of text-to-SQL models, emphasizing correct query generation and abstention from infeasible questions to foster safer deployment.

Contribution

The paper presents TrustSQL, a comprehensive benchmark with a penalty-based scoring metric for assessing and enhancing the reliability of text-to-SQL models, including new evaluation approaches.

Findings

Existing methods struggle under severe penalties, indicating room for improvement.

Unified models can be evaluated effectively with the new penalty-based metric.

Achieving high reliability scores requires significant effort in model development.

Abstract

Text-to-SQL enables users to interact with databases using natural language, simplifying the retrieval and synthesis of information. Despite the remarkable success of large language models (LLMs) in translating natural language questions into SQL queries, widespread deployment remains limited due to two primary challenges. First, the effective use of text-to-SQL models depends on users' understanding of the model's capabilities-the scope of questions the model can correctly answer. Second, the absence of abstention mechanisms can lead to incorrect SQL generation going unnoticed, thereby undermining trust in the model's output. To enable wider deployment, it is crucial to address these challenges in model design and enhance model evaluation to build trust in the model's output. To this end, we introduce TrustSQL, a novel comprehensive benchmark designed to evaluate text-to-SQL reliability-defined as a model's ability to correctly handle any type of input question by generating correct SQL queries for feasible questions and abstaining from generating infeasible ones (e.g., due to schema incompatibility or functionalities beyond SQL). We evaluate existing methods using a novel penalty-based scoring metric with two modeling approaches: (1) pipeline-based methods combining SQL generators with infeasible question detectors and SQL error detectors for abstention; and (2) unified methods using a single model for the entire task. Our experimental results reveal that achieving high scores under severe penalties requires significant effort and provide a new perspective on developing text-to-SQL models for safer deployment. TrustSQL is available at https://github.com/glee4810/TrustSQL.