LorExperts compresses MoE experts by clustering them, keeping one full-precision dominant per cluster, and encoding other experts as permutation-aligned low-rank corrections, preserving the router and outperforming prior compression baselines at ~50% compression.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Shape Mutating Expert Compression:LorExperts and BTExperts
LorExperts compresses MoE experts by clustering them, keeping one full-precision dominant per cluster, and encoding other experts as permutation-aligned low-rank corrections, preserving the router and outperforming prior compression baselines at ~50% compression.