Entity
Mixture-of-Experts (MoE)
A neural network architecture that uses multiple expert networks with a gating mechanism to activate only relevant subsets for each input.
术语属性
- Application：Used in DeepSeek V3 architecture
- Benefit：Efficient inference with large parameter counts