1
TENP: Trapezoidal Expert Neuron Pruning For Mixture-of-Experts
MoE大模型专家剪枝新方法，梯形剪枝保留路由拓扑，大幅压缩静态参数占用。
arXiv:2606.09885v1 Announce Type: new Abstract: Mixture-of-Experts large language models (LLMs) scale efficiently through sparse activation, yet their…