AI Papers Daily

#Teacher Exposure

1 篇

Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning

用于LLM推理自蒸馏的自适应教师暴露

🧭 LLM ★ 深度解读

ByteDance

提出自适应教师暴露控制（ATESD），将教师所见的特权推理长度变为可学习变量，解决教师侧暴露不匹配问题

LLM | 2026-05-11 | HF Daily | score 9/10

Self-DistillationLLMReasoningTeacher ExposureReinforcement Learning