CoRD：耦合 rollout 弥补在线策略蒸馏的盲区
CoRD: Coupled Rollouts Fill the Blind Spot of On-Policy Distillation
训练小模型的人可以看看这篇：CoRD 改进在线策略蒸馏，Qwen3 系列 0.6B 到 4B 上数学基准 pass@1 平均多涨约 1 到 2 个点，收敛还快 5 倍。
arXiv 论文提出 CoRD（coupled-rollout distillation），保留在线策略损失，同时通过 speculative decoding 的最大耦合把训练状态分布向教师模型偏移。在 Qwen3 与 Qwen3.5 两代 0.6B 到 4B 共六个学生模型上，CoRD 在五个公开数学基准上比在线策略蒸馏的 pass@1 平均高 0.9 到 2.0 个点，pass@8 高 1.2 到 3.4 个点。4B 学生模型在教师可达状态上用 16 步达到的差距收敛，在线策略蒸馏 80 步仍未达到。
CoRD: Coupled Rollouts Fill the Blind Spot of On-Policy Distillation
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.