On-Policy Distillation (5 minute read)
Summary
This paper introduces on-policy distillation, which trains a student model on its own trajectories with teacher token-level KL supervision to fix train-inference mismatch, unifying forward-KL, reverse-KL, and JSD losses, with reverse-KL favored for smaller students.
Similar Articles
Demystifying On-Policy Distillation: Roles, Pathologies, and Regulations
This paper systematically studies on-policy distillation in LLM post-training, clarifying its role as an exploration catalyst and identifying pathologies like Student-Teacher Mismatch and Length Exploitation, proposing light-weight signal regulations.
The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes
This paper presents a comprehensive empirical study on on-policy distillation for large language models, identifying failure mechanisms like distribution mismatch and optimization instability, and proposing fixes such as stop-gradient objectives and RLVR-adapted teachers.
When Teachers Mislead: Spurious-Signal-Aware On-Policy Distillation
This paper introduces SA-OPD, a spurious-signal-aware on-policy distillation framework that filters misleading token-level teacher supervision based on input-grounding and optimization impact, improving LLM and VLM distillation performance.
Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement
The paper analyzes on-policy distillation, revealing it primarily suppresses low-probability tokens rather than relying on teacher guidance, and introduces OPSA, a supervision-free method that significantly enhances reasoning performance.
OPRD: On-Policy Representation Distillation
OPRD proposes a new knowledge distillation method that aligns student and teacher hidden states across layers during on-policy rollouts, eliminating sampling variance from token-space KL estimation. Empirically, OPRD outperforms output-space baselines on math reasoning benchmarks (AIME 2024/2025, AIMO) while being 1.44x faster and using 54% less memory.