1 paper · 1 filter
Songming Zhang, Xue Zhang, Tong Zhang +3
Knowledge distillation (KD) is widely used to compress and post-train large language models (LLMs), yet many existing frameworks execute teacher inference with the same training-or…