Reproduce “TranSAGE: Automating Spatial Accelerator Generation for Large-Scale Transformer Models” 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
Chi Zhang, Yi Sun, Weichuang Zhang, Quan Chen, Jieru Zhao
TranSAGE maps dependent Transformer operations into clusters that the compiler schedules as shared hardware pipelines. It searches cluster partitions and hardware mappings to balance energy and delay while limiting resource contention. For server decode at batch size 8, it reports average throughput 3.37× LEGO across four models and sequence lengths 128–4096; this generated-accelerator comparison is model-based.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.