跳到正文
原文
AK(@_akhaliq)· AK (@_akhaliq)·· 2026-09-02AI 评估 · 24/100分数表示内容值得关注的程度,依据实质份量、信息增量、证据强度、读者相关性和可用性,按内容类型加权。显示分数为两次独立评分的平均值。AI 判断可能有误,评分不代表事实已被核实,也不同于热点榜的讨论热度。

论文《Does On-Policy Distillation Really Distill?》探讨从噪声教师到自我改进

Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement paper: https:/...

AI 导读

论文《Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement》已发布在 Hugging Face,探讨 on-policy 蒸馏是否真正实现蒸馏,并提出从噪声教师到自我改进的路径。

来源:AK(@_akhaliq) · x.com