Think2SQL: Bridging the Reasoning Gap in Text-to-SQL for Small LLMs

Leveraging RL with our reward mechanism, we push Qwen-Coder-2.5 7B to performance on par with much larger LLMs (>400B) on the BIRD dataset! 🤯

Model: huggingface.co/simone-papic...
Paper: huggingface.co/papers/2504....

Details 👇

A huge thanks to @madelonhulsebos.bsky.social and the entire organizing team for putting together such an engaging and insightful workshop.

There were great discussions, valuable perspectives, and an inspiring research community. I look forward to the next one!✨

3/3 🧵

🔍 In our previous work, in collaboration with @papotti.bsky.social and Luca Cagliero, we introduced DAMBER (Data-Ambiguity Tester)—a tool designed to systematically analyze and evaluate ambiguous queries. Exciting developments are on the way; stay tuned for our follow-up project! 🚀

2/3 🧵

🚀 Had an amazing time at the @ellisamsterdam.bsky.social for the Workshop on Representation Learning and Generative Models for Structured Data, presenting how ambiguity impacts performance in Text2SQL evaluation 😶‍🌫️

1/3 🧵