Mixture-of-Experts (MoE) has become a popular technique for scaling large language models (LLMs) without exploding computational costs. Instead of using the entire model capacity for every input, MoE ...

Scientific publishing relies on peer review as the mechanism that maintains trust in what we publish. When we read a journal article, we assume experts have rigorously scrutinized it before ...