Is DeepSeekMoE superseded?

DeepSeekMoE (Mixture-of-experts routing): heavily superseded — a standard baseline that newer methods routinely beat. 0 paper(s) critique it, 6 beat it on benchmarks — #4 of 1370 most-superseded. Sub-problem: cluster led by DeepSeekMoE. Newer alternatives in the same sub-problem include ProbMoE, Grouter, DECO, SPHERE, AIR-MoE.

Mixture-of-experts routing

DeepSeekMoE

DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models

Heavily superseded#4 of 1,370 most-superseded · first seen Jan 11, 2024

Heavily superseded — a standard baseline that newer methods routinely beat

0 papers critique it · 6 beat it on benchmarks

Beaten on benchmarks

Head-to-head results where a newer method reports beating DeepSeekMoE. Values are copied from the source paper's tables — verify against the cited paper.

DS-MoE-6B beats DeepSeekMoE
4603.9 vs 3144.1
TPS · [H100-80GB GPU]
Dense Training, Sparse Inference: Rethinking Training of Mixture-of-Experts Language Models
BlockFFN beats DeepSeekMoE
71.38 vs 49.27
CLS_8 · [Small]
BlockFFN: Towards End-Side Acceleration-Friendly Mixture-of-Experts with Chunk-Level Activation Sparsity
SPHERE beats DeepSeekMoE
0.45 vs 0.33
average · [MetaWorld DS-MoE CRL]
SPHERE: Mitigating the Loss of Spectral Plasticity in Mixture-of-Experts for Deep Reinforcement Learning
Mistral-MoCE beats DeepSeekMoE
55.03 vs 40.77
Avg · [Cross-model comparison]
Mixture-of-Clustered-Experts: Advancing Expert Specialization and Generalization in Instruction Tuning
IDA-MoE beats DeepSeekMoE
43.1 vs 38.4
VizWiz · [StableLM-1.6B + CLIP-336]
Input Domain Aware MoE: Decoupling Routing Decisions from Task Optimization in Mixture of Experts
OmniMoE beats DeepSeekMoE
50.9 vs 50.2
Avg · [6.4B-A1.7B models]
OmniMoE: An Efficient MoE by Orchestrating Atomic Experts at Scale

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.