Reproduce “An FPGA-Based Reconfigurable Accelerator for Convolution-Transformer Hybrid EfficientViT” 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

Haikuo Shao, Huihong Shi, Wendong Mao, Zhongfeng Wang

The accelerator maps EfficientViT’s mixed operations to a reconfigurable processing element and a second compute engine, scheduling convolution and attention work in parallel. Its time-multiplexed, pipelined dataflow supports depthwise and pointwise convolutions and lightweight attention. On a ZCU102 FPGA at 200 MHz, it reports over 95% utilization for EfficientViT-B1, with a first-convolution exception, and 780.2 GOPS, but no traffic measurement.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.