Reproduce “HEMERA: A Heterogeneous Memory-Centric Accelerator with Recursive Dataflow for Edge-Constrained State-Space-Duality Models 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
Hao Ding, Ling Liang, Ruitong Qiao, Dongxue Zhao, Xiantong Qiu, Jinshan Li, Meng Li, Lei Jin and 4 more
HEMERA changes Mamba-2’s Structured State Space Duality (SSD) execution by streaming state updates as a recurrence rather than constructing pairwise token interactions. It preserves the computation; digital compute-in-memory units handle dense projections. For the 2.8B model, HEMERA reports 1.4× average latency speedup over the optimized fused A100 kernel; evaluation uses RTL-measured compute latency and an analytical memory model.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.