Reproduce “Turbo-GMM: FlashAttention-Inspired GPU-Accelerated Gaussian Mixture Model” 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
Momin Abbas, Sadia Asif, Asad Ul Haq, Kaoutar El Maghraoui
Turbo-GMM fits diagonal-covariance Gaussian mixture models on GPUs by keeping all components on chip for a data tile, normalizing scores and accumulating update statistics before leaving. This removes a second component traversal when the component axis fits; otherwise, it uses two passes. On one A100, it achieves a 2.15× median speedup over Flash-GMM across 15 completed comparisons, including an invalid-fit result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.