rapidsai/cuml

Implements 40+ classical ML algorithms (clustering, regression, dimensionality reduction, time series) with scikit-learn-compatible APIs, enabling drop-in GPU acceleration without CUDA expertise. Leverages CUDA kernels and libraries like Faiss for 10-50x speedups on tabular data, with multi-GPU/multi-node support via Dask and UCXX for distributed training and inference across clusters.

5,143 stars. Actively maintained with 72 commits in the last 30 days.