MoE

An architecture that balances parameter count and compute efficiency by sparsely activating multiple expert sub-networks.

No public content is connected to this entity yet.