Reproduce “VAQF: Fully Automatic Software-Hardware Co-Design Framework for Low-Bit Vision Transformer” 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
Mengshu Sun, Haoyu Ma, Guoliang Kang, Yifan Jiang, Tianlong Chen, Xiaolong Ma, Zhangyang Wang, Yanzhi Wang
VAQF automatically chooses activation precision and FPGA accelerator settings for a target frame rate. On DeiT-base on a Xilinx ZCU102, its 6-bit design reached 31.6 FPS against a 30-FPS target, with 76.5% ImageNet-1K top-1 accuracy; the 8-bit design reached 24.8 FPS against 24 FPS, with 77.6% accuracy. These findings are limited to the tested setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.