A framework that skips unimportant layers and exits early during edge MoE inference to cut latency and cross-server traffic, with quality kept within a budget.
A survey of large language models,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.NI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
TrimMoE A communication aware and adaptive depth framework for distributed edge inference
A framework that skips unimportant layers and exits early during edge MoE inference to cut latency and cross-server traffic, with quality kept within a budget.