1 paper · 1 filter
Md Nurul Absar Siddiky
Sparse mixture-of-experts (MoE) language models activate only a small subset of parameters for each token, making router behavior a central part of model computation. This paper st…