Understanding On-Policy Distillation: A Mechanistic Interpretability Perspective via Sparse Crosscoders

arXiv:2609.35210 · cs.CL · Submitted 2026-09-28 · Read on arXiv

cs.CL

Submitted: 2026-09-28

Updated: 2026-09-28

Project page: https://yzc-666.github.io/understanding-opd-crosscoders/ABSTRACT

Terminology

• Distillation — 19× in this paper · explained in 2 episodes
• On-Policy Distillation — 16× in this paper · explained in 3 episodes
• Accuracy — 10× in this paper · explained in 1 episode
• Trace — 10× in this paper · explained in 2 episodes
• Benchmark — 6× in this paper · explained in 2 episodes
• GRPO — 5× in this paper · explained in 6 episodes
• Adam — 4× in this paper · explained in 2 episodes
• Transformer — 4× in this paper · explained in 2 episodes
• Alignment — 3× in this paper · explained in 1 episode
• Failure Modes — 3× in this paper · explained in 3 episodes
• Reinforcement Learning — 3× in this paper · explained in 4 episodes
• Sparsity — 3× in this paper · explained in 4 episodes
• Interpretability — 2× in this paper · explained in 15 episodes
• Verification — 2× in this paper · explained in 2 episodes
• Adaptation — 1× in this paper · explained in 1 episode
• Agentic AI — 1× in this paper · explained in 1 episode
• Decode — 1× in this paper · explained in 2 episodes
• Exposure Bias — 1× in this paper · explained in 2 episodes
• Fidelity — 1× in this paper · explained in 2 episodes
• Generative AI — 1× in this paper · explained in 2 episodes
• Knowledge Distillation — 1× in this paper · explained in 11 episodes
• Reasoning Traces — 1× in this paper · explained in 3 episodes
• Regret — 1× in this paper · explained in 3 episodes
• Self-distillation — 1× in this paper · explained in 6 episodes
• Test-time Scaling — 1× in this paper · explained in 2 episodes

Sources

• BatchTopK Sparse Autoencoders
• Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement
• Towards Understanding On-Policy Distillation through the Lens of Test-Time Scaling
• OpenThoughts: Data Recipes for Reasoning Models
• DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
• JustRL: Scaling a 1.5B LLM with a Simple RL Recipe
• Skywork Open Reasoner 1 Technical Report
• LoRA: Low-Rank Adaptation of Large Language Models
• Cross-Architecture Model Diffing with Crosscoders: Unsupervised Discovery of Differences Between LLMs
• Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe
• Simple-OPD: Demystifying Warm-up for On-policy Distillation
• Your Teacher Can't Help You Here: Combating Supervision Fidelity Decay in On-Policy Distillation
• Demystifying OPD: Length Inflation and Stabilization Strategies for Large Language Models
• HybridFlow: A Flexible and Efficient RLHF Framework
• Kimi K3: Open Frontier Intelligence
• MiMo-V2-Flash Technical Report
• Escaping the KL Agreement Trap in On-Policy Distillation
• Qwen3 Technical Report
• Mismatch Matters: On-Policy Distillation Beyond Token Agreement
• GLM-5: from Vibe Coding to Agentic Engineering

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering