The authors developed CB-Orchestrator as an alternative to using one fixed workflow for an entire task or generating a new workflow for every query.
Contextual bandits select reusable LLM workflows with less training overhead
CB-Orchestrator builds a workflow pool through evolutionary search, then routes each query to a suitable workflow using a sample-efficient contextual bandit.
Chinese Tech
Jiahang Sun · Zhiwei Shang · Zhipiao Liu · Hongwei Yang · Guoqing Xie · Shuang Qiu · +1 more
The Chinese University of Hong Kong, Shenzhen · Huawei Technologies · City University of Hong Kong
Research Digest··2 min read
Sun et al.
Why this paper
From Huawei Technologies and 2 others
In one line
CB-Orchestrator beats baselines on five benchmarks and cuts MATH training token overhead by over 40% via contextual bandit workflow selection.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§