菜单
Disagreement-Modulated Policy Self-Distillation framework for LLM reasoning. Resolves privileged information leakage and exploration preservation in on-policy distillation via reverse-KL barycenter target, achieving leakage attenuation and exploration preservation simultaneously.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/hiyenwong/ai_collection --skill demopsd-policy-self-distillation命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Graph-native Python reimplementation of the Information Dynamics of Music (IDyOM) model that represents predictive memories as explicit graph objects for musical expectation modeling and network analysis.
Physics-aware end-to-end deep reinforcement learning methodology for quadcopter control with actuator dynamics modeling.
Reinforced Dreamer methodology for asymmetric reinforcement learning using latent guidance to improve world model representations and behaviors in model-based RL.
基于 SOC 职业分类
正在显示 SKILL.md
| name | demopsd-policy-self-distillation |
| description | Disagreement-Modulated Policy Self-Distillation framework for LLM reasoning. Resolves privileged information leakage and exploration preservation in on-policy distillation via reverse-KL barycenter target, achieving leakage attenuation and exploration preservation simultaneously. |
| trigger | policy self-distillation, DemoPSD, privileged information leakage, reverse KL, knowledge distillation, LLM reasoning training |
| category | ai_collection/collection/skills |
DemoPSD is a novel on-policy self-distillation framework that resolves two fundamental problems in LLM reasoning distillation:
Instead of fitting the full teacher distribution, DemoPSD steers the student toward a reverse-KL barycenter target — a weighted geometric combination of teacher and student distributions.
# Pseudo-implementation
teacher_dist = model(prompt, privileged=True)
student_dist = model(prompt, privileged=False)
disagreement = kl_divergence(teacher_dist, student_dist)
barycenter = geometric_mean(teacher_dist, student_dist, weight=disagreement)
loss = reverse_kl(student_dist, barycenter)
policy self-distillation, DemoPSD, privileged information leakage, reverse KL, knowledge distillation, exploration preservation
arXiv: 2607.02497 (2026-07-02)