PIVOT: Perplexity-Informed KD-to-RL Transition Scheduling for Vertical-Domain Few-Shot Distillation

Researchers proposed PIVOT (Perplexity-Informed Transition Optimization), a dynamic framework that routes samples between On-Policy Distillation and GRPO reinforcement learning based on teacher-evaluated sequence perplexity, according to arXiv paper 2610.11167v1. On the Banking77 and HWU64 benchmarks, PIVOT outperformed continued OPD and globally synchronized OPD-to-GRPO baselines under the same number of post-warm-up student optimization steps, yielding stronger downstream performance and more stable training dynamics. The method addresses vertical-domain few-shot classification for small language models by moving low-perplexity samples to GRPO for reward-driven refinement while keeping high-perplexity samples under OPD for continued domain knowledge acquisition.

arXiv:2610.11167v1 Announce Type: new Abstract: Vertical-domain few-shot classification remains challenging for small language models, as limited supervision makes it difficult to acquire domain-specific decision knowledge. On-Policy Distillation OPD can improve teacher-guided adaptation by supervising student-generated rollouts, while GRPO-based reinforcement learning can further refine downstream predictions. However, existing KD-to-RL pipelines typically rely on globally fixed transition schedules, ignoring that different samples may require different amounts of teacher-guided acquisition before reward-driven refinement. We propose PIVOT Perplexity-Informed Transition Optimization , a dynamic transition framework that routes samples between OPD and GRPO according to teacher-evaluated sequence perplexity. PIVOT moves low-perplexity samples to GRPO for reward-driven refinement while keeping high-perplexity samples under OPD for continued domain knowledge acquisition. Experiments on Banking77 and HWU64 show that PIVOT consistently outperforms continued OPD and globally synchronized OPD$\rightarrow$GRPO baselines under the same number of post-warm-up student optimization steps, achieving stronger downstream performance and more stable training dynamics.