Agentic On-Policy Distillation Implicitly Reshapes LLMs' ReAct Patterns
DeyangKong, Li Zhaoyi, Sijian Ren, Yuan Wei, Ying Wei, Defu Lian, Shikun Zhang, Wei Ye
Agentic On-Policy Distillation Implicitly Reshapes LLMs' ReAct Patterns: 7 upvotes on Hugging Face Daily Papers, #20 of 39 papers on 2026-11-18. Day-by-day upvote history.
Upvotes over time
It had 7 upvotes by its first day, 7 upvotes by the next day, 7 upvotes by day 3, 7 upvotes by day 7, and has 7 now. It ranked #61 of 146 papers in its week and #283 of 520 in its month.
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 τ^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 τ^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 τ^2 accuracy from 11.5% to 29.9% with a source-learning trade-off.
Paper page on Hugging Face · arXiv
Data: hysts-bot-data/daily-papers-stats and the Daily Papers API. Open data: tardellirs/paper-pulse-data. Sister project: Model Pulse, the download history of every model on the Hub.