The authors first studied agents trained through supervised fine-tuning on expert trajectories composed of repeated thought, action and observation steps.
Progress-aware training helps small language model agents escape repetitive loops
Across eight models, the authors report that penalizing stalled trajectories improves task success and completion efficiency.
Big Tech
Yuanpu Cao · Saket Sathe · Hanyu Wang · Ziyi Yin · Fenglong Ma · Jinghui Chen
The Pennsylvania State University · Microsoft CoreAI
Research Digest··3 min read
Cao et al.
Why this paper
From Microsoft CoreAI and The Pennsylvania State University
In one line
ProD detects and penalizes stagnation in small language model agents, improving task success and efficiency.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§