The authors treat requesting cloud assistance as an action available directly to the small model, rather than assigning the decision to a separate routing system.
Small agents learn when to seek cloud help
Sibyl trains a 0.6-billion-parameter model to request step-level assistance only when useful, then absorb that guidance into later decisions.
Chinese Tech
Zhewei Fang · Yuxin Zhang · Zhenwei Shao · Mengze Li · Zheng Lin · Long Chen · +5 more
Hangzhou Dianzi University · Fudan University · The Hong Kong University of Science and Technology · University of Luxembourg · Simon Fraser University
Research Digest··3 min read
Fang and colleagues present Sibyl, a three-stage reinforcement-learning framework for combining an on-device small language model with a stronger cloud model on tasks requiring many sequential actions.
Why this paper
From Alibaba Group and 5 others
In one line
Sibyl achieves 95.2% success on ALFWorld and 80.4% on WebShop, averaging 0.8 and 3.9 cloud calls per trajectory.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (params 0.6B)
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§