DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models
논문 링크 DeepSeekMoE 정리 – Dense 성능을 40 % FLOPs로 끌어낸 ‘세분화-전문가’ 트릭한 줄 요약 (TL;DR)Fine-Grained Expert Segmentation (FGES) + Shared Experts (SEI) 로 …
40분
None
None
1페이지