single mention · model architecture
Hash routing
Hash routing is an expert routing strategy for MoE layers that determines which experts each token goes to.1 It uses a predefined hash function of the input token ID instead of a learned router.2
How this definition was written
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
The Hash routing strategy determines the target experts of each token
— DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligenceaccording to a predefined hash function with regard to the input token ID
— DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
Filed under model architecture › channel mixer › mixture of experts › expert routing