CSGO: Generalized Optimization for Cold Start in Wireless Collaborative Edge LLM Systems
Xuran Liu, Nan Xue, Rui Bao, Yaping Sun, Zhiyong Chen, Meixia Tao, Xiaodong Xu, Shuguang Cui · 2025
Read CSGO: Generalized Optimization for Cold Start in Wireless Collaborative Edge LLM Systems by Xuran Liu, Nan Xue, Rui Bao, Yaping Sun, Zhiyong Chen, Meixia Tao, Xiaodong Xu, Shuguang Cui (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-11287.txt.
Description
Xuran Liu and co-authors cut the startup lag of large language models split across wireless edge devices, in a 2025 paper. Running a big model on small gear near the user promises fast, private answers, but the gear can hold only slices of it. Slices load on demand, and that cold-start loading leaves the pipeline idle. Past splitting schemes largely ignored the wait.
Their framework, CSGO, overlaps loading with computing and message passing to hide the dead time. It tunes how layers are cut and which device takes which slice, based on each device's strength and the model's shape. The puzzle is cast as a mixed-integer program, solved by a lean dynamic programming method. Tests show cold-start lag well below the base strategies. The authors pitch the scheme as a step toward smooth team inference at the edge.
Catalog record
| CREATOR | Xuran Liu, Nan Xue, Rui Bao, Yaping Sun, Zhiyong Chen, Meixia Tao, Xiaodong Xu, Shuguang Cui |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-11287