RAD: Redundancy-Aware Distillation for Hybrid Models via Self-Speculative Decoding

Open in new window