---
title: "CoRD: Coupled Rollouts Fill the Blind Spot of On-Policy Distillation"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2611.07959v1"
permalink: "https://feed.astralor.com/items/itm_4b2377428d9fd4f81a32cc79"
published: "2026-11-05T12:55:51.000Z"
score: 58
channel: "ai"
tags: ["RAG","CoRD","Coupled Rollouts Fill","Blind Spot","On-Policy Distillation On-policy","On-policy"]
---

# CoRD: Coupled Rollouts Fill the Blind Spot of On-Policy Distillation

## 速览

On-policy distillation trains a student on its own rollouts, so states the teacher would reach but the student does not are seldom supervised. We introduce CoRD (coupled-rollout distillation), which keeps the on-policy …

## 关键点

- On-policy distillation trains a student on its own rollouts, so states the teacher would reach but the student does not are seldom supervised.
- We introduce CoRD (coupled-rollout distillation), which keeps the on-policy loss and moves only the distribution of training states toward the teacher, drawing rollouts from a geometric interpolation of student and teac…
- The coupling keeps the student's token wherever the two distributions agree and otherwise substitutes a token from the part of the target the student underweights, so interventions are as rare as any exact sampler allow…

## 全文

On-policy distillation trains a student on its own rollouts, so states the teacher would reach but the student does not are seldom supervised. We introduce CoRD (coupled-rollout distillation), which keeps the on-policy loss and moves only the distribution of training states toward the teacher, drawing rollouts from a geometric interpolation of student and teacher through the maximal coupling of speculative decoding. The coupling keeps the student's token wherever the two distributions agree and otherwise substitutes a token from the part of the target the student underweights, so interventions are as rare as any exact sampler allows and recede as the student learns. Across six base students from two model families (Qwen3 and Qwen3.5, 0.6B to 4B), CoRD raises average pass@1 over on-policy distillation by 0.9 to 2.0 points and pass@8 by 1.2 to 3.4 points on five public math benchmarks at equal steps and data. On frozen teacher-reached states, a 4B student trained with CoRD reaches in 16 steps a gap closure that on-policy distillation does not reach in 80, and substitution rules intervening at least as often in other directions fall short of CoRD.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_4b2377428d9fd4f81a32cc79