פרופ' גיא כץ
האוניברסיטה העברית בירושלים
Accelerating Program Verification with Small Language Models
Abstract
The synthesis of inductive loop invariants remains a critical bottleneck in automated program verification. While Large Language Models (LLMs) show promise in mitigating this issue, they often fail on complex programs, producing invariants that are invalid or computationally ineffective. We present Wonda, a pipeline for fine-tuning Small Language Models (SLMs) to enable the to produce high-quality invariants. On the challenging InvBench suite, our approach enables small models to outperform off-the-shelf models 20x their size.
Based on joint work with Ido Pinto, Yizhak Yisrael Elboher, Haoze Wu and Nina Narodytska. This work appeared in ICML 2026.








