OVD: On-policy Verbal Distillation
This addresses memory bottlenecks and exploration limitations in on-policy distillation for efficient student models, offering a novel method for specific bottlenecks in reinforcement learning.
The paper tackles the problem of knowledge distillation in reinforcement learning by introducing On-policy Verbal Distillation (OVD), which replaces token-level probability matching with trajectory matching using verbal scores, resulting in up to +12.9% absolute improvement in EM on Web Q&A and up to +25.7% gain on math benchmarks.
Knowledge distillation offers a promising path to transfer reasoning capabilities from large teacher models to efficient student models; however, existing token-level on-policy distillation methods require token-level alignment between the student and teacher models, which restricts the student model's exploration ability, prevent effective use of interactive environment feedback, and suffer from severe memory bottlenecks in reinforcement learning. We introduce On-policy Verbal Distillation (OVD), a memory-efficient framework that replaces token-level probability matching with trajectory matching using discrete verbal scores (0--9) from teacher models. OVD dramatically reduces memory consumption while enabling on-policy distillation from teacher models with verbal feedback, and avoids token-level alignment, allowing the student model to freely explore the output space. Extensive experiments on Web question answering and mathematical reasoning tasks show that OVD substantially outperforms existing methods, delivering up to +12.9% absolute improvement in average EM on Web Q&A tasks and a up to +25.7% gain on math benchmarks (when trained with only one random samples), while also exhibiting superior training efficiency. Our project page is available at https://OVD.github.io