From Proprietary to Open-Source: Bridging the Distribution Gap via Multi-Agent Protocol Distillation in Agentic Search
2026-07-27 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors created a method called Multi-Agent Protocol Distillation (MAPD) to help AI language models learn better by combining step-by-step problem solving with retrieving information. Instead of just copying answers or styles from more advanced, private models, their method uses a structured 'protocol' that breaks down reasoning into clear parts like plans and facts. This approach gives more detailed guidance during training and improves the model’s ability to solve questions on various tests. Their results showed that MAPD works better than existing techniques and helps keep the AI focused, avoiding unnecessary style changes.
agentic searchreinforcement learningknowledge distillationmulti-agent systemprotocol distillationreasoning planextractive groundingQA benchmarkslogit matchingstyle drift
Authors
Junlin Liu, Jiangwang Chen, Zixin Song, Shuaiyu Zhou, Chunji Lv, Hank Wu, Kailin Jiang, Jinyang Wu, Bohan Yu, Chenxi Zhou
Abstract
Agentic search enables large language models to solve knowledge-intensive tasks by interleaving multi-step reasoning with retrieval, yet optimizing this with outcome-based reinforcement learning (RL) provides only sparse supervision. Knowledge distillation can supply denser guidance, and advanced proprietary models with their strong reasoning capabilities are promising teachers. While distilling from proprietary models can densify this supervisory signal, conventional logit-matching is precluded by hidden logits and mismatched tokenizers, whereas raw natural language trajectory imitation transfers superficial stylistic artifacts rather than core reasoning competence. To address the heterogeneous distillation problem and bridge the distribution gap, we propose Multi-Agent Protocol Distillation (MAPD), a joint distillation and RL framework uses a structured, style-normalized protocol as an intermediate representation. An offline multi-agent system (MAS) decomposes each query, retrieves supporting evidence, repairs failed searches, and converts the resulting exploration trace into a JSON protocol containing the task type, reasoning plan, and extractive grounding facts. During training, the protocol is provided only to a privileged branch of the student policy, whose token distributions furnish a dense distillation signal alongside the sparse RL objective. Extensive evaluations across seven QA benchmarks demonstrate that MAPD consistently outperforms competitive distillation and RL, achieving average success rates of 39.4\% on Qwen3-1.7B and 44.4\% on Qwen3-4B. Crucially, the framework generalizes robustly across diverse proprietary teachers while effectively mitigating the student policy from style drift and verbosity degeneration.