Public KG record
Mixture-of-Experts
Recorded overview
An architecture where only a subset of experts (K) is active per token, but all
An evidence-backed entity from the project knowledge graph; detailed tool metadata is still being added.
Recorded details
- Capabilities: 路由器为每个 token 选取少量专家网络进行处理
- Capabilities: Dynamic top-k
- Capabilities: Expert choice routing
- Capabilities: Group-limited and node-limited routing
- Capabilities: Hash routing
- Capabilities: Shared plus routed experts
- Capabilities: Sigmoid gating with aux-loss-free bias
- Capabilities: Softmax top-k token choice
Evidence coverage
The KG currently records 9 known information dimensions and has 10 undocumented dimensions, linked to 39 public source articles. These counts describe record coverage, not model performance or verified fit.
Related public records
- FreeToken — based on
- Grouped matmul — applied to
- HOPE — applied to
- HOPE — improves
- Humming — improves