PreMoE uses Predicted Expert Utility scores to compile sparse MoE variants achieving up to 50% sparsity with nearly no performance loss on models from 30B to 718B parameters.
This pruned set of experts becomes the new set of routed experts{E r i (x)} M i=1 for the compiled instance
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
PreMoE: Proactive Inference for Efficient Mixture-of-Experts
PreMoE uses Predicted Expert Utility scores to compile sparse MoE variants achieving up to 50% sparsity with nearly no performance loss on models from 30B to 718B parameters.