libcuml-cu12

What it is and what it does

libcuml-cu12 is the CUDA 12 variant of cuML, NVIDIA's GPU-accelerated machine learning library that mirrors scikit-learn's API for tabular ML tasks. It implements clustering, regression, classification, dimensionality reduction, time series, and preprocessing—all running on NVIDIA GPUs instead of CPU.

The library is designed for data scientists and engineers who want to accelerate traditional ML workflows without writing CUDA code. It depends on cuda-toolkit, libcuvs-cu12, libraft-cu12, librmm-cu12, nvidia-nvjitlink-cu12, and rapids-logger. Multi-GPU and multi-node operation is available via Dask for a growing set of algorithms. Models can be serialized with pickle or joblib for later inference, though deserialization from untrusted sources poses security risks.

Use it for:

• Accelerate large-scale clustering tasks on GPU when CPU processing is too slow.
• Train and deploy regression or classification models on GPU with scikit-learn-compatible interface.
• Perform distributed nearest neighbors search across multiple GPUs using Dask.
• Run dimensionality reduction on large tabular datasets to prepare features for downstream tasks.
• Build time series forecasts on GPU-resident data without CPU bottlenecks.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

GPU-accelerated machine learning algorithms with scikit-learn-compatible APIs for clustering, regression, classification, dimensionality reduction, and time series models.

Yes, if you have NVIDIA GPU hardware and need to accelerate scikit-learn-style ML workflows. The active maintenance, permissive Apache-2.0 license, and scikit-learn API compatibility make it a low-risk choice. Install friction is moderate due to CUDA and GPU library dependencies, but that is inherent to GPU compute; no known vulnerabilities. Not suitable without GPU hardware.

Install

libcuml-cu12 on PyPI

pip install libcuml-cu12

uv add libcuml-cu12

poetry add libcuml-cu12

Installing libcuml-cu12

Before you install

Medium install friction due to CUDA toolkit and multiple NVIDIA GPU libraries (libcuvs-cu12, libraft-cu12, librmm-cu12, nvidia-nvjitlink-cu12) as runtime dependencies. Actively maintained with release 8 days old; requires Python 3.11 or higher.

License in practice

Apache-2.0 permissive license allows commercial and private use with minimal restrictions; suitable for proprietary projects provided attribution and license text are included.

Quickstart

pip install libcuml-cu12

from libcuml-cu12 import cluster

# GPU-accelerated clustering with DBSCAN
dbscan = cluster.DBSCAN(eps=1.0, min_samples=1)
dbscan.fit(data)
labels = dbscan.labels_

Requires NVIDIA GPU with CUDA support and cuda-toolkit installed; Python 3.11 minimum.

Verify before relying

• Performance claim of 10-50x speedup versus CPU equivalents—exact conditions and dataset sizes not specified.
• Compatibility with scikit-learn version 1.6 or higher—whether this is a hard requirement or advisory.
• Multi-GPU and multi-node support scope—which algorithms fully support Dask distributed operation.
• Exact import paths and module structure for clustering, regression, and other algorithm categories.

Package facts

Evidence: libcuml_cu12-26.8.0-py3-none-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl; libcuml_cu12-26.8.0-py3-none-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl

Tags

More Scientific/Engineering packages

Further reading