PulseAugur
实时 11:02:53
English(EN) Distill Skills into Weights, Not Prompts: Abstract Skills as Privileged Signals for On-Policy Self-Distillation

新框架统一策略内自蒸馏以增强LLM推理能力 · 跟踪3个来源

研究人员开发了一个统一的策略内自蒸馏(OPSD)框架,通过将特权信息整合到模型参数中来增强LLM的推理能力。这个名为统一策略内自蒸馏(USD)的新框架,通过根据学生模型的学习能力联合优化token选择和特权信息调整,解决了先前OPSD方法的局限性。另一种名为SKALD(Skill-Anchored Latent Distillation)的方法使用抽象技能卡提供密集监督,尤其是在标准奖励信息不足时,在数学基准测试上显示出显著的改进。另一项独立研究OP2SD,探讨了上下文诱导的教师行为在OPSD中的作用,表明收益不仅来自访问参考解决方案,还来自教师上下文的变化。 AI

影响 这些在自蒸馏方面的研究进展可能带来更高效、更强大的LLM,以应对复杂的推理任务。

排序理由 该集群包含三篇在arXiv上发表的学术论文,详细介绍了与大型语言模型的策略内自蒸馏相关的新方法和分析。

在 Hugging Face Daily Papers 阅读 →

AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →

新框架统一策略内自蒸馏以增强LLM推理能力 · 跟踪3个来源

报道来源 [4]

  1. arXiv cs.AI TIER_1 English(EN) · Yongkang Yang, Zhezheng Hao, Hong Zhang, Yi Liu, Xiankun Lin, Wence Ji, Fanjunduo Wei, Jiarui Yu, Qiang Lin, Xiaoyun Liang, Hande Dong ·

    匹配监督与学生学习能力:一种单策略自蒸馏的统一框架

    arXiv:2608.08176v1 Announce Type: new Abstract: On-policy self-distillation (OPSD) improves the reasoning abilities of LLMs by internalizing privileged context into model parameters through self-distillation. Two recent research lines promote vanilla OPSD by choosing which tokens…

  2. arXiv cs.AI TIER_1 English(EN) · Yuki Ichihara, Naoto Iwase, Mohammad Atif Quamar, Junpei Komiyama ·

    特权解决方案还是情境诱导的教师行为?深入剖析 on-policy 自蒸馏

    arXiv:2608.09228v1 Announce Type: cross Abstract: On-Policy Self-Distillation (OPSD) is commonly interpreted as the transfer of privileged information: a teacher observes the verified solution to the target problem and supervises the student's trajectory. However, this interpreta…

  3. arXiv cs.AI TIER_1 English(EN) · Yubo Jiang, Fengying Xie, Zhiguo Jiang, Haopeng Zhang ·

    将技能提炼为权重而非提示:将抽象技能作为策略内自蒸馏的特权信号

    arXiv:2608.09826v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards yields no group-relative signal when rollout groups are uniformly correct or uniformly wrong, which account for 63.0-68.0% of groups in our experiments. We propose SKALD (Skill-Anchor…

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

    将技能提炼为权重而非提示:将抽象技能作为策略内自蒸馏的特权信号

    Reinforcement learning with verifiable rewards yields no group-relative signal when rollout groups are uniformly correct or uniformly wrong, which account for 63.0-68.0% of groups in our experiments. We propose SKALD (Skill-Anchored Latent Distillation), an on-policy self-distill…