Rethinking Self-Distillation for Multi-Teacher Capability Merging

arXiv:2610.04272 · cs.LG, cs.AI, cs.CL · Submitted 2026-10-03 · Read on arXiv

cs.LG, cs.AI, cs.CL

Submitted: 2026-10-03

Updated: 2026-10-03

Code: https://github.com/apple-aiml-research/ml-rethink-mopd

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

Terminology

• Distillation — 1× in this paper · explained in 2 episodes
• Self-distillation — 1× in this paper · explained in 6 episodes

Sources

• On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes
• Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks
• REGEN: Replay-recycling for Expert-to-Generalist distillation with Offline Reinforcement Learning
• Training Verifiers to Solve Math Word Problems
• Model Breadcrumbs: Scaling Multi-Task Model Merging with Sparse Masks
• DELLA-Merging: Reducing Interference in Model Merging through Magnitude-Based Sampling
• DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
• Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement
• Show Your Work: Improved Reporting of Experimental Results
• RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignment
• Group-in-Group Policy Optimization for LLM Agent Training
• GLM-5: from Vibe Coding to Agentic Engineering
• Measuring Mathematical Problem Solving With the MATH Dataset
• Distilling the Knowledge in a Neural Network
• What Matters in On-Policy Distillation? A Perspective on Data Efficiency and Data Selection
• Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters
• Editing Models with Task Arithmetic
• Model Stock: All we need is just a few fine-tuned models
• Sequence-Level Knowledge 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: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks