IOLM-DB compiles recurring semantic operators over table columns into specialized models that are 1.8 to 2.8 times smaller in memory than the general-purpose model. Our semantic cache for OLAP canonicalizes SQL and natural-language queries with an LLM and reaches an 82% hit rate with zero false hits.
Research Goal
Language models as efficient database operators
