Models & ArchitectureUpdated 2026-10-23
Mixture of Experts
Splitting a model into expert subnetworks and activating only a subset per input.
#模型架构#MoE#成本
In plain language
A router selects experts, keeping per-token computation manageable even with a large total parameter count. MoE balances cost and capability but complicates training and serving, and expert load can be uneven.
Example
A 100B-parameter model may activate only 10B per token, with inference cost near a 10B model.
Recommended tools
Related terms
Fundamentals
Inference
The computation a trained model performs to produce output from an input.
3 related terms
FundamentalsFeatured
Large Language Model
A model trained on massive text to understand and generate natural language.
3 related terms
Models & Architecture
Quantization
Representing model weights at lower precision to reduce memory and inference cost.
3 related terms
Applications
Agentic Workflow
Breaking a task into steps planned by AI, executed by tools and reviewed by humans.
3 related terms