Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination

About

Multi-agent LLM systems have shown promise for complex reasoning, yet recent evaluations reveal they often underperform single-model baselines. We identify a structural failure mode in sequential fine-tuning of shared-context teams: updating one agent shifts the team's context distribution, and when subsequent updates are evaluated on cached rollouts, this mismatch compounds. We formalize this as the compounding occupancy shift and prove that stale-occupancy evaluation incurs a penalty that scales quadratically with the number of agents. In contrast, intermediate-occupancy evaluation reduces this to linear scaling. We propose TeamTR, a trust-region framework that resamples trajectories after each component update and enforces per-agent divergence control, yielding rigorous per-update and per-stage improvement lower bounds. Experiments show that TeamTR outperforms single-agent and sequential baselines with 7.1% on average, mitigates coordination regressions, and supports plug-and-play component replacement. Code is available at https://github.com/Yydc/TeamTR.

Yi Xie, Siao Liu, Falong Fan, Yuanqi Yao, Yue Zhao, Bo Liu• 2026

Related benchmarks

TaskDatasetResultRank
PlanningPlanBench
Accuracy44.5
47
General ReasoningZebraLogic
Accuracy (%)96.3
29
General ReasoningAutoLogi
Accuracy90.5
29
Active ReasoningAR-Bench-DC
Exact Accuracy46.7
23
Mathematical ReasoningMATH 500
Accuracy99.3
21
Active ReasoningARBench-GN
Accuracy45.3
13
General ReasoningAIME 2024
Accuracy (AIME 2024)89.7
13
General ReasoningAIME 2025
Accuracy (%)80.1
13
Active ReasoningARBench SP
Accuracy43.1
13
Showing 9 of 9 rows

Other info

Follow for update