Reproduce “A Pipelined FPGA Architecture for Banded Sparse Matrix Dense Matrix Multiplication in Longformer” 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
Phillip Pramberger, Athanasios Tziouvaras, Shreejith Shanker, George Floros
The authors present a design for multiplying Longformer’s banded attention matrix by a dense value matrix. It stores rows in fixed diagonal-offset order, avoiding explicit column indices, and pipelines two computation paths. Functional simulation reports one dot-product output per clock after 11 startup cycles, with peak throughput above 100 million outputs per second at 100 megahertz, not an end-to-end inference measurement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.