PulseAugur
中
实时 13:17:36
English(EN) Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement

新的 OPSA 方法质疑 on-policy distillation 的有效性

作者 PulseAugur 编辑部 · [2 个来源] ·

研究人员质疑了 on-policy distillation (OPD) 在大型语言模型中的有效性，发现其监督可能存在噪声，并且学生模型对这种噪声基本不敏感。OPD 的收益似乎源于抑制低概率 token，而非直接的教师指导。这促使了 On-Policy Self-Adaptation (OPSA) 的开发，这是一种新的无监督方法，利用熵自适应负优势来提高模型性能。OPSA 显著提升了推理能力，在 AIME24 等基准测试中表现优于 OPD。 AI

影响 提出了一种更有效、无监督的改进 LLM 推理的方法，可能减少对噪声教师模型的依赖。

排序理由 提出新方法并分析现有方法的学术论文。

在 Hugging Face Daily Papers 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →

新的 OPSA 方法质疑 on-policy distillation 的有效性

本文如何被排名

Signal score
2 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Research
提出新方法并分析现有方法的学术论文。
Source corroboration
2 independent sources
Multiple independent publishers reporting the same story raises confidence that it's real and newsworthy.
Topics
paper, model release
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
1 days old
Coverage has settled into its steady-state source set.

完整方法见我们的编辑标准。

报道来源 [2]

  1. arXiv cs.CL TIER_1 English(EN) · Yi Ding, Ruqi Zhang ·

    On-Policy Distillation 真的能蒸馏吗？从噪声教师到自我改进

    arXiv:2608.31046v1 Announce Type: cross Abstract: On-policy distillation (OPD) offers dense token-level supervision as an alternative to the sparse outcome-level advantages of reinforcement learning with verifiable rewards (RLVR). However, the teacher scores student-generated tra…

  2. Hugging Face Daily Papers TIER_1 English(EN) ·

    On-Policy Distillation 真的能蒸馏吗？从噪声教师到自我改进

    On-policy distillation relies mainly on suppressing low-probability tokens rather than teacher guidance, motivating a supervision-free entropy-adaptive method that substantially improves reasoning performance.

相关实体

相关话题