EXAONE Demand 1.0 Foundation Model for Demand Forecasting

EXAONE Demand utilizes a demand-specific corpus of 11.3M series and a demand-aware adapter to handle short histories and frequent zeros. The architecture uses low-rank branches for four demand classes—smooth, intermittent, erratic, and lumpy—selected via a statistical router.