single mention · model architecture
Disabled router-bias updates
An implementation setting that disables router-bias updates during quantization-aware distillation.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See expert load balancing for the methods filed beside it.
Also called Disable router-bias updates during QAD.
Filed under model architecture › channel mixer › mixture of experts › expert load balancing