STEPS: Selective On-Policy Self-Distillation for Reasoning

arXiv:2605.10194 · cs.AI, cs.LG · Submitted 2026-05-11 · Read on arXiv

cs.AI, cs.LG

Submitted: 2026-05-11

Updated: 2026-09-30

Terminology

• Trace — 4× in this paper · explained in 2 episodes
• GRPO — 3× in this paper · explained in 6 episodes
• Distillation — 2× in this paper · explained in 2 episodes
• Self-Distillation — 2× in this paper · explained in 7 episodes
• Alignment — 1× in this paper · explained in 1 episode
• Benchmark — 1× in this paper · explained in 2 episodes
• Information Leakage — 1× in this paper · explained in 2 episodes
• Reinforcement Learning — 1× in this paper · explained in 4 episodes
• Verifiable Rewards — 1× in this paper · explained in 2 episodes

Sources

• On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes
• Learning to Steer, Steering to See: Unveiling the Geometry of RLVR in Large Language Models via Trainable Vectors
• DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
• GLM-5: from Vibe Coding to Agentic Engineering
• OpenThoughts: Data Recipes for Reasoning Models
• Distilling the Knowledge in a Neural Network
• Reinforcement Learning via Self-Distillation
• The Art of Scaling Reinforcement Learning Compute for LLMs
• Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?
• Unifying Group-Relative and Self-Distillation Policy Optimization via Sample Routing
• Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe
• Filter, Then Reweight: Rethinking Optimization Granularity in On-Policy Distillation
• Critical Tokens Matter: Token-Level Contrastive Estimation Enhances LLM's Reasoning Capability
• Self-Distilled Policy Gradient
• MiMo-V2-Flash Technical Report
• Learning Query-Specific Rubrics from Human Preferences for DeepResearch Report Generation
• Privileged Information Distillation for Language Models
• GPQA: A Graduate-Level Google-Proof Q&A Benchmark
• DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
• Self-Distillation Enables Continual Learning

Related papers

• MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
• Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
• Agents' Last Exam
• The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
• MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
• Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems