✨
AI Summary
- Mixture-of-Recursions (MoR) enhances large language model efficiency.
- Combines parameter sharing and adaptive computation with recursive Transformers.
- Uses lightweight routers for dynamic recursion depth and efficient KV caching.