#190190th of 242 in ML frameworks and libraries

gpytorch

Gaussian process library built on PyTorch with GPU-accelerated inference

Stars
3.9k
License
MIT
Last commit
Oct 2026
Last release
Feb 2026
Language
Python
Stars, last 29 weeks+78
25 Apr 2026: 3,860 stars2 May 2026: 3,860 stars9 May 2026: 3,865 stars16 May 2026: 3,867 stars23 May 2026: 3,872 stars30 May 2026: 3,877 stars6 Jun 2026: 3,880 stars13 Jun 2026: 3,884 stars20 Jun 2026: 3,889 stars27 Jun 2026: 3,890 stars4 Jul 2026: 3,891 stars11 Jul 2026: 3,894 stars18 Jul 2026: 3,896 stars25 Jul 2026: 3,897 stars1 Aug 2026: 3,899 stars8 Aug 2026: 3,900 stars15 Aug 2026: 3,905 stars22 Aug 2026: 3,905 stars29 Aug 2026: 3,906 stars5 Sept 2026: 3,907 stars12 Sept 2026: 3,909 stars19 Sept 2026: 3,913 stars26 Sept 2026: 3,918 stars3 Oct 2026: 3,923 stars10 Oct 2026: 3,924 stars17 Oct 2026: 3,926 stars24 Oct 2026: 3,931 stars31 Oct 2026: 3,932 stars7 Nov 2026: 3,937 stars14 Nov 2026: 3,938 stars

3,860 on 25 Apr 20263,938 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

GPyTorch is a Python library for building Gaussian process models on PyTorch. Most inference uses numerical linear algebra such as preconditioned conjugate gradients, via the LinearOperator interface, instead of Cholesky decomposition. It includes implementations of SKI/KISS-GP, stochastic Lanczos expansions, LOVE, SKIP, stochastic variational GPs and deep kernel learning.

Who it is for: ML researchers building scalable Gaussian process models in PyTorch

Strengths

  • Matrix-multiplication-based inference makes better use of GPUs than Cholesky solvers
  • Implements SKI/KISS-GP, LOVE, SKIP, stochastic variational GPs and deep kernel learning
  • New scalable methods can be added by supplying a kernel matrix multiply via LinearOperator
  • Integrates with PyTorch models; optional KeOps and Pyro extras

Weaknesses

  • Requires Python 3.10+, PyTorch 2.4.1+ and NumPy 2.0+
  • Library only: no UI, server or hosted service
  • Iterative solvers add approximation and tuning concerns versus exact Cholesky GPs
  • Focused on Gaussian processes; not a general-purpose probabilistic modeling toolkit

What it needs

  • GPU optional
  • Needs PyTorch >= 2.4.1, NumPy >= 2.0, linear_operator
  • Models: Gaussian processes, SKI/KISS-GP, stochastic variational GP, deep kernel learning

What changed

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in ML frameworks and libraries

See all 242
Also in ML frameworks and libraries
RankProjectScore
1transformersModel-definition library for pretrained text, vision, audio and multimodal models169.3k starsApache-2.094 out of 100
2ultralyticsPython library and CLI for training and running YOLO vision models63.5k starsAGPL-3.094 out of 100
3pytorch-image-modelsPyTorch library of image models, pretrained weights and training scripts37.3k starsApache-2.094 out of 100
#4pytorchPython tensor library with GPU support and tape-based autograd105.8k starscustom license93 out of 100
#5supervisionPython toolkit for loading, annotating and counting computer vision detections Live demo ↗ (opens in a new tab)51.7k starsMIT92 out of 100