#214214th of 242 in ML frameworks and libraries

hummingbird

Compiles trained scikit-learn, LightGBM and XGBoost models into PyTorch tensor ops

Stars
3.6k
License
MIT
Last commit
Sep 2026
Last release
Oct 2024
Language
Python
Stars, last 29 weeks+37
25 Apr 2026: 3,517 stars2 May 2026: 3,518 stars9 May 2026: 3,519 stars16 May 2026: 3,522 stars23 May 2026: 3,522 stars30 May 2026: 3,522 stars6 Jun 2026: 3,523 stars13 Jun 2026: 3,524 stars20 Jun 2026: 3,524 stars27 Jun 2026: 3,524 stars4 Jul 2026: 3,525 stars11 Jul 2026: 3,526 stars18 Jul 2026: 3,526 stars25 Jul 2026: 3,528 stars1 Aug 2026: 3,530 stars8 Aug 2026: 3,533 stars15 Aug 2026: 3,536 stars22 Aug 2026: 3,536 stars29 Aug 2026: 3,537 stars5 Sept 2026: 3,537 stars12 Sept 2026: 3,538 stars19 Sept 2026: 3,539 stars26 Sept 2026: 3,540 stars3 Oct 2026: 3,542 stars10 Oct 2026: 3,546 stars17 Oct 2026: 3,550 stars24 Oct 2026: 3,550 stars31 Oct 2026: 3,551 stars7 Nov 2026: 3,552 stars14 Nov 2026: 3,554 stars

3,517 on 25 Apr 20263,554 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

Hummingbird converts trained traditional ML models, such as scikit-learn decision trees and random forests and LightGBM and XGBoost classifiers and regressors, into tensor computations. Targets are PyTorch, TorchScript, ONNX and TVM, so the models can run on GPU and be served with TorchServe. It keeps a scikit-learn-style inference API, so a converted model can replace the original with one convert() call.

Who it is for: Engineers speeding up tree-model inference on GPU or PyTorch stacks

Strengths

  • One convert(model, 'pytorch') call keeps the sklearn-style predict API unchanged
  • Converted tree models run on GPU via model.to('cuda')
  • Exports to PyTorch, TorchScript, ONNX and TVM
  • Compilation strategies described in OSDI 2020 and PVLDB 2021 papers

Weaknesses

  • Supports a fixed operator set; other backends and models are only on the roadmap
  • Last release was 2024-10-24; tested on Python 3.9 to 3.11 only
  • TVM backend only works through Python 3.10
  • PyTorch >= 1.6.0 must be installed separately

What it needs

  • GPU optional
  • Docker
  • Needs PyTorch >= 1.6.0, scikit-learn, LightGBM (optional), XGBoost (optional)
  • Models: scikit-learn Decision Trees and Random Forest, LightGBM, XGBoost, PyTorch, TorchScript

What changed

  1. “In https://github.com/microsoft/hummingbird/pull/761 users are now required to add an additional flag if they wish to unzip without the checksum hashes. Ex: model.load(location, overrideflag=True) . …”

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