牛哥精选 · 三个月

📋 全部 🤖 AI·大模型 ⚡ 效率工具 📝 深度技术 🚀 产品观察 💰 商业科技 🔓 开源项目 🎨 设计创意 📖 阅读推荐 🏷 资源合集 🌱 成长效率

📝 深度技术 arXiv 机器学习 2026-05-21

Complementing reinforcement learning with SFT through logit averaging in the post training of LLMs

新方法通过logit averaging融合强化学习与监督微调，显著提升LLM后训练的性能和稳定性。

arXiv:2605.20555v1 Announce Type: new Abstract: We introduce a novel method that averages the logits of a frozen reference policy (e.g., SFT) and a tr…

大语言模型强化学习监督微调 logit aver 后训练

🤖 AI·大模型 arXiv 机器学习 2026-05-20

Decoupling KL and Trajectories: A Unified Perspective for SFT, DAgger, Offline RL, and OPD in LLM Distillation

一篇统一SFT、DAgger、离线RL和OPD视角的LLM蒸馏论文，解耦KL与轨迹，为模型优化提供新理论框架。

arXiv:2605.16826v1 Announce Type: new Abstract: Knowledge distillation is central to LLM post-training, yet its design space remains poorly understood…

llm蒸馏 kl散度监督微调强化学习轨迹优化

📅 日期

2026-05-20 2026-05-19

🐂 牛哥精选

Complementing reinforcement learning with SFT through logit averaging in the post training of LLMs

Decoupling KL and Trajectories: A Unified Perspective for SFT, DAgger, Offline RL, and OPD in LLM Distillation

📅 日期