When EOS Tokens Disagree: Understanding Length Inflation in On-Policy Distillation

arXiv:2609.20511 · cs.LG · Submitted 2026-09-17 · Read on arXiv

cs.LG

Submitted: 2026-09-17

Updated: 2026-09-17

Comments: 30 pages, 12 figures, 3 tables, code available at https://github.com/UNCSciML/opd-eos

Code: https://github.com/UNCSciML/opd-eos

Project page: https://uncsciml.github.io/opd-eos-website

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Distillation — 51× in this paper · explained in 2 episodes
• On-Policy Distillation — 20× in this paper · explained in 3 episodes
• Aggregation — 12× in this paper · explained in 1 episode
• Alignment — 12× in this paper · explained in 1 episode
• Action Space — 7× in this paper · explained in 1 episode
• Accuracy — 5× in this paper · explained in 1 episode
• Benchmark — 5× in this paper · explained in 2 episodes
• Ablation — 4× in this paper · explained in 1 episode
• Fine-Tuning — 2× in this paper · explained in 5 episodes
• Knowledge Distillation — 2× in this paper · explained in 11 episodes
• Verification — 2× in this paper · explained in 2 episodes
• Actor — 1× in this paper · explained in 3 episodes
• Adam — 1× in this paper · explained in 2 episodes
• Credit assignment — 1× in this paper · explained in 3 episodes
• Decoupling — 1× in this paper · explained in 2 episodes
• Distribution Shift — 1× in this paper · explained in 6 episodes
• Entropy Collapse — 1× in this paper · explained in 2 episodes
• Reinforcement Learning — 1× in this paper · explained in 4 episodes
• Trace — 1× in this paper · explained in 2 episodes
• Verifier — 1× in this paper · explained in 2 episodes

Sources

• Revisiting On-Policy Distillation: Empirical Failure Modes and Simple Fixes
• The Llama 3 Herd of Models
• Entropy-Aware On-Policy Distillation of Language Models
• Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe
• ADWIN: Adaptive Windows for Horizon-Aware On-Policy Distillation
• Prefix Teach, Suffix Fade: Local Teachability Collapse in Strong-to-Weak On-Policy Distillation
• Demystifying OPD: Length Inflation and Stabilization Strategies for Large Language Models
• Trust-Region Behavior Blending for On-Policy Distillation
• Gemma 3 Technical Report
• Demystifying On-Policy Distillation: Roles, Pathologies, and Regulations
• On the Position Bias of On-Policy Distillation
• Pass the Baton: Trajectory-Relayed On-Policy Distillation
• Qwen3 Technical Report
• Learning beyond Teacher: Generalized On-Policy Distillation with Reward Extrapolation
• Prune-OPD: Efficient and Reliable On-Policy Distillation for Long-Horizon Reasoning
• Decoupling KL and Trajectories: A Unified Perspective for SFT, DAgger, Offline RL, and OPD in LLM Distillation

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks