cs.AINov 14, 2026

Agentic On-Policy Distillation Implicitly Reshapes LLMs' ReAct Patterns

Authors: Deyang Kong, Zhaoyi Li, Sijian Ren, Yuan Wei, Ying Wei, Defu Lian, Shikun Zhang, Wei Ye

Organizations: Peking University · University of Science and Technology of China · Tsinghua University · Zhejiang University

Abstract

On-policy distillation (OPD) transfers specialized capabilities to language-model agents, but its cross-domain effects on interaction remain unclear. We investigate how OPD reshapes ReAct interaction patterns across mathematics, ALFWorld, and τ2τ^2-bench, focusing on reasoning, actions, and observation acquisition. General-reasoning supervision reduces mathematical tool attempts from 70.4% to 14.4% while increasing accuracy from 16.7% to 38.0%; code-interpreter supervision improves household actions and retail queries. In contrast, ALFWorld supervision reduces τ2τ^2-bench accuracy from 41.5% to 6.6%, accompanied by frequent tool calls that violate user-information dependencies. To address negative transfer using only source-domain training trajectories, we explore selective OPD guided by teacher preference changes before and after domain-specific RL. Retaining the top 95% of positions raises τ2τ^2 accuracy from 11.5% to 29.9% with a source-learning trade-off.

Figures & tables

Explore similar work

CardsList
  1. Multi-Turn On-Policy Distillation with Prefix Replay

    Jul 6, 2026Baohao Liao, Hanze Dong, Christof Monz +3LLM Agent TrainingOn-Policy Distillation

  2. ActionOPD: Distilling Decisions, Not Tokens

    Oct 14, 2026Huining Yuan, Xiangmin Yi, Zelai Xu +5LLM Agent TrainingKnowledge Distillation

  3. Every Coin Has Two Sides: On the Dual Nature of Generalization in On-Policy Distillation of Large Language Models

    Aug 17, 2026Zhaoyi Li, Deyang Kong, Yuan Wei +13Language Model DistillationOn-Policy Distillation