mixture-of-recursions

MoR (Mixture-of-Recursions) is an innovative architecture that improves LLM efficiency, achieving up to 2x faster inference speeds without accuracy loss…