Inference Optimisation

Mixture-of-Experts Inference

Why serving MoE models is harder than serving dense models of equivalent quality, and how DeepSeek and Mistral made it work in production.

advanced · 9 min read · Premium