Reproduce “CoFormer: Collaborating with Heterogeneous Edge Devices for Scalable Transformer Inference” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Guanyu Xu, Zhiwei Hao, Li Shen, Yong Luo, Fuhui Sun, Xiaoyan Wang, Han Hu, Yonggang Wen
CoFormer enables transformer inference across heterogeneous edge devices by decomposing a pretrained model into smaller models that run in parallel and aggregating their intermediate features centrally. Its search uses device profiles and a validation-loss proxy to select decompositions, followed by calibration. On ImageNet-1K, it reports up to 3.1× faster classification than Swin-L, with a 1.7% accuracy sacrifice in this comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.