You Only Train Once: 더 효율적으로 Kanana SLM 개발하기 | AI 방향성
0:00 · Watch on YouTube
Kakao's Kana SLM team built 3B/1.3B/0.9B on-device models via iterative pruning + distillation, then moved to Kana Elastic Training to train multiple nested sizes in one run with a learned router instead of repeated c...