Knowledge Distillation for Low-Resource Open-source Text-to-SQL Model

In three linesKnowledge distillation framework for Text-to-SQL in low-resource settings. Constructs knowledge base (schema semantics, abbreviations, business logic) injected during training and inference. Generates contextually grounded synthetic data. Evaluated on 7 benchmarks: improves open-source and closed-source LLMs, especially on domain-specific datasets.

Summary generated by Claude — human-verified