arXiv:2506.13329 EAQuant: Enhancing PTQ for MoE Models via Expert-Aware Optimization

Routing-instability-is-the-real-problem framing: activation outliers, routing instability, and sparse expert calibration are named as the three failures of generic PTQ on MoE. Introduces routing consistency alignment to preserve expert selection post-quantization. Reports average accuracy gains of 1.15-13.81% at W4A4/W3A4/W3A3/W2A4 across three MoE architectures.