arXiv cs.CL· Lin Yao·· 10 小时前AI 评分24
多教师在线策略蒸馏的域路由增量优化（DRD-MOPD）
Domain-Routed Delta Optimization for Multi-Teacher On-Policy Distillation
AI 导读
研究者提出 Domain-Routed Delta Optimization（DRD-MOPD），为多教师在线策略蒸馏中的每个领域样本指定应更新哪些参数组件，学生模型由冻结基座与可训练组件相加构成。
正文 · 原文
Abstract:Multi-teacher on-policy distillation assigns each sample a domain teacher but typically updates shared student parameters. We introduce Domain-Routed Delta Optimization (DRD-MOPD), which specifies which parameter components each domain's samples update. The student is the sum of a frozen base and trainable components initialized from expert-minus-base weights. All components participate in every forward pass; four update modes define the correspondence between sample domains and components, each with independent optimizer states. We analyze these update rules and measure retention of the original expert geometry through angles, projection coefficients, and energy. Prior work such as Open-MOPD distills directly from the base model; 97.02% of its released final checkpoint's increment energy lies outside the span of the original expert directions. In comparison, our Own checkpoint after 20 updates retains coefficients near one along each component's original direction and pairwise angles ranging from 88.04 to 92.41 degrees; its exported BF16 student's increment has 4.90% outside energy, versus 10.52% for the sum of saved FP32 components. The same checkpoint achieves a 32.72% macro score across six benchmarks, compared with 32.44% for its initial task-arithmetic merge. These measurements describe geometry and capability at the evaluated training stages and settings.
| Subjects: | Machine Learning (cs.LG); Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.13249 [cs.LG] |
| (or arXiv:2611.13249v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2611.13249 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Lin Yao [view email]
[v1]
Mon, 9 Nov 2026 21:21:41 UTC (75 KB)
来源：arXiv cs.CL · arxiv.org