Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

PADD: Path-Aligned Decompression Distillation for Non-Router Teacher to Guide MoE Student Learning

About

As large language models (LLMs) continue to scale, it becomes increasingly challenging to grow model capacity under fixed computation budgets. We propose Path-Aligned Decompression Distillation (PADD), a framework for distilling knowledge from dense teachers without explicit routing into mixture-of-experts (MoE) students while learning high-quality routing policies. PADD organizes knowledge distillation into four stages in two phases: an initialization phase (Stage I) that builds diverse functionality in the student's experts through teacher neuron clustering and student-expert warmup, and a training phase (Stages II--IV) that integrates online adaptive distillation, path-refined policy optimization, and reward-augmented load balancing in a single training pipeline. Experiments on mathematical reasoning benchmarks demonstrate that PADD yields substantial gains over strong baselines at the same inference cost and that the MoE student can match or surpass its dense teacher. They also demonstrate effective teacher-to-student knowledge distillation and stable routing behavior.

Xinyue Peng, Yi Qian, Jiaojiao Lin, Wenjian Shao, Yanming Liu• 2026

Related benchmarks

TaskDatasetResultRank
Mathematical ReasoningMATH 500
Top-1 Accuracy96.4
452
Mathematical ReasoningOlympiadBench
Pass@1 Accuracy70.7
16
Mathematical ReasoningAIME24
Pass@1 Accuracy83
16
Showing 3 of 3 rows

Other info

Follow for update