Train multiple small LoRAs (dialogue, description, philosophy, translation) and route per prompt using a tiny classifier (or even heuristics). This improves specialization without retraining a giant monolith.
· Sign up or log in to comment
