الملخص
On-policy distillation (OPD) reduces train-test state mismatch by training a student on its own generated trajectories, but weak students may visit teacher-misaligned prefixes where supervision is less representative. We introduce SAKI (Supervision Allocation with KL-constrained Interpolation), which combines a KL-constrained teacher-guided rollout with maximal coupling and reuses realized accept/correction events to route token-level supervision. Accepted positions retain sampled-token reverse-KL supervision, while correction positions receive direct supervision on the teacher's highest-probability token. Under maximal coupling, the correction probability is exactly TV(p_t, q_t), so the same trust-region radius controls rollout deviation and upper-bounds intervention and specialized-supervision frequency. We further implement an engine-resident speculative verifier that preserves the exact-q trajectory distribution and coupling semantics while improving matched-workload rollout throughput by 4.22x. Across seven mathematical reasoning benchmarks, SAKI improves the matched teacher-guided baseline in Mean@8 and Pass@8 for both 1.7B and 0.6B students. Placement controls and fixed-prefix analysis further support correction-triggered routing as a conflict-adaptive supervision signal.
الكلمات المفتاحية
الموضوع
بيانات النشر
- المجلة
- غير متاح
- وصول مفتوح
- وصول مفتوح أخضر
اقتبس هذه المقالة
APA 7
Wei, M., Wang, X., Chen, Z., Chai, J., Liu, S., Wang, L., Xu, H., Hu, Z., Lin, W., & Yin, G. (2026). SAKI: Maximal-Coupling-Routed Teacher Supervision for On-Policy Distillation. https://omanscience.com/ar/articles/saki-maximal-coupling-routed-teacher-supervision-for-on-policy-distillation
MLA 9
Wei, Miteto, et al. "SAKI: Maximal-Coupling-Routed Teacher Supervision for On-Policy Distillation." https://omanscience.com/ar/articles/saki-maximal-coupling-routed-teacher-supervision-for-on-policy-distillation.
شيكاغو (المؤلف–التاريخ)
Wei, Miteto, Xiaohan Wang, Zehao Chen, Jiajun Chai, Sichao Liu, Li Wang, Haoyuan Xu, Zhaoyu Hu, Wei Lin, and Guojun Yin. 2026. "SAKI: Maximal-Coupling-Routed Teacher Supervision for On-Policy Distillation." https://omanscience.com/ar/articles/saki-maximal-coupling-routed-teacher-supervision-for-on-policy-distillation.
هارفارد
Wei, M., Wang, X., Chen, Z., Chai, J., Liu, S., Wang, L., Xu, H., Hu, Z., Lin, W. and Yin, G. (2026) 'SAKI: Maximal-Coupling-Routed Teacher Supervision for On-Policy Distillation', Available at: https://omanscience.com/ar/articles/saki-maximal-coupling-routed-teacher-supervision-for-on-policy-distillation.
فانكوفر
Wei M, Wang X, Chen Z, Chai J, Liu S, Wang L, et al. SAKI: Maximal-Coupling-Routed Teacher Supervision for On-Policy Distillation. https://omanscience.com/ar/articles/saki-maximal-coupling-routed-teacher-supervision-for-on-policy-distillation
IEEE
M. Wei, X. Wang, Z. Chen, J. Chai, S. Liu, L. Wang, H. Xu, Z. Hu, W. Lin, and G. Yin, "SAKI: Maximal-Coupling-Routed Teacher Supervision for On-Policy Distillation," https://omanscience.com/ar/articles/saki-maximal-coupling-routed-teacher-supervision-for-on-policy-distillation.