Mixture-of-Depths: Dynamically allocating compute in transformer-based language models Paper • 2404.02258 • Published Apr 2 • 104
