Speculate, then Collaborate: Fusing Knowledge of Language Models during Decoding

Open in new window