List your company

Text Summarization on CNN/DM (TPS, Latency)

50.17TPSMoE-SpAc
-0.68612.51725.7238.923Feb 12, 2026

Evaluation results

MethodLinks
MoE-SpAcBase Model=DeepSeek-V2-Lite2026.0250.178.13
llama.cpp-w/ SDBase Model=DeepSeek-V2-Lite, Speculative Decoding=true2026.0233.2315.94
MoE-SpAc2026.0222.7814.31
HybriMoEBase Model=DeepSeek-V2-Lite2026.0218.4321.78
llama.cppspeculative decoding=true2026.0216.4618.36
llama.cppspeculative decoding=false2026.0215.4219.46
HybriMoE2026.0212.3725.11
Fate2026.027.8340.78
SP-MoE2026.024.3865.24
MoE-Infinity2026.023.5680.56
vLLM2026.022.57104.28
Mixtral Offload2026.022.23131.72
Accelerate2026.021.27222.47