Search papers, labs, and topics across Lattice.
The paper introduces R$^3$-SQL, a Text-to-SQL framework that improves ranking consistency and candidate recall. It addresses inconsistent scoring of functionally equivalent SQL queries by grouping candidates by execution result and using a combined pairwise preference and pointwise utility to rank these groups. To improve recall, R$^3$-SQL employs agentic resampling, selectively generating new candidates when the initial pool is deemed insufficient. R$^3$-SQL achieves state-of-the-art execution accuracy on the BIRD-dev benchmark, demonstrating consistent gains across five benchmarks.
Text-to-SQL models can now achieve significantly higher accuracy by grouping and ranking SQL candidates based on execution results, then strategically resampling when the initial pool is lacking.
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.
