arXiv：cs.CL· Wenxi Gan·· 3 小时前AI 评分38
从紧凑教师监督中蒸馏小型智能体：Task-Progress Distillation 方法
Learning to Act with Task Progress: Distilling Small Agents from Compact Teacher Supervision
AI 导读
研究者提出 Task-Progress Distillation（TPD），将示范动作与当前任务阶段标签配对，让小型智能体通过联合评分阶段-动作对来选择动作。在 ALFWorld 上，1.7B 学生模型用 404 条示范达到 72.4% 未见任务平均成功率，而推理训练的学生仅 48.3%。
正文
Abstract:Learning from large-model demonstrations offers a way to train small agents that can complete recurring tasks without calling a large model at every step. A central design choice is what to retain from teacher trajectories that contain reasoning, actions, and information about task progress. We introduce Task-Progress Distillation (TPD), an offline approach that pairs each demonstrated action with a short label describing the current task stage. The student learns these compact targets and selects actions by jointly scoring admissible stage--action pairs, which a deterministic harness executes in the environment. On ALFWorld, a 1.7B student trained with 404 demonstrations achieves 72.4\% mean unseen task success with either TPD or action-only supervision, compared with 48.3\% for a reasoning-trained student using constrained action selection. Explicit stages provide an additional benefit at 200 demonstrations, improving success from 48.0\% to 67.7\% over action-only supervision. With more demonstrations, the action-only student closes the gap, and both approaches reach 76.9\% at 808 demonstrations. Shared-history analyses link part of TPD's local advantage to better decisions when moving between subgoals, particularly from object acquisition to processing. These results show that compact supervision can train effective small task agents, while explicit task progress provides additional guidance at an intermediate demonstration budget.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.10332 [cs.CL] |
| (or arXiv:2610.10332v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.10332 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Wenxi Gan [view email]
[v1]
Wed, 7 Oct 2026 16:15:56 UTC (1,451 KB)
来源：arXiv：cs.CL · arxiv.org