Empirical benchmarks show MoE inference cost on edge hardware tracks total parameters rather than active parameters, with OLMoE-1B-7B behind dense baselines especially on the Jetson device.
Tsinghua Science and Technology31(3), 1365–1380 (2026)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.PF 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
Does Mixture-of-Experts Actually Help Inference on Consumer and Edge Hardware? An Empirical Study
Empirical benchmarks show MoE inference cost on edge hardware tracks total parameters rather than active parameters, with OLMoE-1B-7B behind dense baselines especially on the Jetson device.