arXiv：cs.LG· Yuanpu Cao, Saket Sathe, Hanyu Wang, Ziyi Yin, Fenglong Ma, Jinghui Chen·· 3 小时前
ProD：面向小语言模型智能体进展停滞的进展感知蒸馏
Progress-Aware Distillation for Mitigating Stagnation in Small Language Model Agents
AI 导读
针对小语言模型（SLM）智能体在长交互轨迹中被蒸馏后易陷入失败策略循环的"进展停滞"问题，研究者提出进展感知蒸馏（ProD），将学生与教师的停滞差距编码为自适应 margin 动态惩罚停滞。在 8 个 0.6B 至 8B 参数的 SLM 上、域内与域外基准测试显示，ProD 显著缓解进展停滞，同时提升蒸馏 SLM 智能体的成功率和任务完成效率。该工作已被 NeurIPS 2026 接收。
正文
Abstract:While LLM-based agent systems have demonstrated remarkable proficiency in reasoning and solving complex real-world tasks, they also come with substantial token consumption and inference latency, limiting their practical deployment. To enhance the efficiency of agent systems, recent efforts have focused on transferring agentic capabilities from large-scale models to small language models (SLMs) via supervised fine-tuning-based distillation. Nevertheless, as interaction trajectories grow longer, involving extended reasoning-action chains and extensive environmental feedback, distilled student agents tend to experience progress stagnation: becoming trapped in unproductive loops characterized by persistent failed strategies and difficulty making meaningful progress. Our systematic analysis shows that this phenomenon consistently occurs across diverse SLMs and task domains. To address this issue, we propose Progress-Aware Distillation (ProD), which explicitly detects and penalizes stagnation in student-generated trajectories. ProD encodes the stagnation gap between student and teacher into an adaptive margin for dynamically penalizing possible stagnation. By iteratively applying this procedure, ProD enables student agents to progressively approximate teacher behavior. Extensive experiments involving 8 SLMs, ranging from 0.6B to 8B parameters, on both in-domain and out-of-domain benchmarks, demonstrate that ProD substantially mitigates progress stagnation while enhancing both the success rates and task-completion efficiency of distilled SLM agents.
| Comments: | Accepted to NeurIPS 2026 |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.19439 [cs.LG] |
| (or arXiv:2610.19439v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.19439 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yuanpu Cao [view email]
[v1]
Wed, 14 Oct 2026 02:31:56 UTC (1,151 KB)
来源：arXiv：cs.LG · arxiv.org