Method Drift›Mixture-of-experts routing
PMQ
Mixture-of-experts routing
superseded — cited as a baseline and beaten by newer methods
1 papers critique it · 3 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites PMQ as a baseline.
“existing approaches remain constrained to local, layer-wise bit allocation and thus fail to capture variations in expert importance across layers”
— GEMQ: Global Expert-Level Mixed-Precision Quantization for MoE LLMs
Beaten on benchmarks
Head-to-head results where a newer method reports beating PMQ. Values are copied from the source paper's tables — verify against the cited paper.
- EAC-MoE: Expert-Selection Aware Compressor for Mixture-of-Experts Large Language Models
QESC beats PMQ · 0-shot^8 [Mixtral-8x7B, 2.06 bits]
66.31 vs 63.25
- EAC-MoE: Expert-Selection Aware Compressor for Mixture-of-Experts Large Language Models
QESC beats PMQ · 0-shot^8 [Mixtral-8x7B, 2.54 bits]
69.61 vs 67.5
- EAC-MoE: Expert-Selection Aware Compressor for Mixture-of-Experts Large Language Models
QESC beats PMQ · 0-shot^8 [Phi3.5-moe, 2.06 bits]
65.03 vs 61.35
- EAC-MoE: Expert-Selection Aware Compressor for Mixture-of-Experts Large Language Models
QESC beats PMQ · 0-shot^8 [Phi3.5-moe, 2.54 bits]
66.53 vs 66.03
- EAC-MoE: Expert-Selection Aware Compressor for Mixture-of-Experts Large Language Models
QESC beats PMQ · 0-shot^8 [Deepseek-moe-16b-base, 2.06 bits]
57.05 vs 54.79
- EAC-MoE: Expert-Selection Aware Compressor for Mixture-of-Experts Large Language Models
QESC beats PMQ · 0-shot^8 [Deepseek-moe-16b-base, 2.54 bits]
58.33 vs 58
- EAC-MoE: Expert-Selection Aware Compressor for Mixture-of-Experts Large Language Models
QESC beats PMQ · 0-shot^8 [Qwen1.5-MoE-A2.7B, 2.06 bits]
59.52 vs 57.79
- EAC-MoE: Expert-Selection Aware Compressor for Mixture-of-Experts Large Language Models
QESC beats PMQ · 0-shot^8 [Qwen1.5-MoE-A2.7B, 2.54 bits]
61.47 vs 60.47
- Efficient Quantization of Mixture-of-Experts with Theoretical Generalization Guarantees
Router norm + Max var (Ours) beats PMQ · Avg. [Mixtral 8x7B, 2.5 bits/expert]
68.38 vs 67.53
- GEMQ: Global Expert-Level Mixed-Precision Quantization for MoE LLMs
GEMQ beats PMQ · 0-shot accuracy [Mixtral-8x7B at 2.5 bpe]
65.13 vs 64.34
- GEMQ: Global Expert-Level Mixed-Precision Quantization for MoE LLMs
GEMQ beats PMQ · WikiText2 perplexity [Mixtral-8x7B at 2.5 bpe]
5.03 vs 5.10
- GEMQ: Global Expert-Level Mixed-Precision Quantization for MoE LLMs
GEMQ beats PMQ · C4 perplexity [Mixtral-8x7B at 2.5 bpe]
9.02 vs 9.21
Newer alternatives
Recent methods in the same sub-problem, not yet superseded in the knowledge base.
- May 22, 2026
- May 21, 2026
- KBVQ-MoEKBVQ-MoE: KLT-guided SVD with Bias-Corrected Vector Quantization for MoE Large Language ModelsJan 30, 2026
- Oct 13, 2025