#121121st of 168 in ML frameworks and libraries
captum
Model interpretability library for PyTorch with attribution algorithms
- Stars
- 5.7k
- License
- BSD-3-Clause
- Last commit
- Oct 2026
- Last release
- Apr 2026
- Language
- Python
5,584 on 21 Apr 20265,721 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Captum is a Python library that explains PyTorch model predictions. It implements attribution methods such as Integrated Gradients, DeepLift, GradientShap, SmoothGrad and VarGrad via NoiseTunnel, layer and neuron conductance, TCAV and TracIn influence functions, plus adversarial attack and perturbation tools. It installs with pip and requires Python 3.10+ and PyTorch 2.3+.
Who it is for: ML researchers and engineers debugging or explaining PyTorch models
Strengths
- Wide set of attribution methods: IG, DeepLift, GradientShap, TCAV, TracIn, conductance
- Works on any PyTorch nn.Module, with layer- and neuron-level attribution
- Convergence delta gives a checkable approximation error for several methods
- Plain pip install, BSD-3-Clause, with tutorials and paper references per algorithm
Weaknesses
- PyTorch only; no support for other frameworks
- DeepLift supports only the Rescale rule and not all non-linear activations
- Captum Insights visualization UI was retired after v0.8.0
- A research library, not an app; users write Python to apply it
What it needs
- GPU optional
- Needs Python >= 3.10, PyTorch >= 2.3
- Models: PyTorch models, torchvision, torchtext
Quick answers
Is captum open source?
Yes. Its code is on GitHub under the BSD-3-Clause license.
Does captum need a GPU?
No. It runs on a CPU and can use a GPU when you have one.
Also in ML frameworks and libraries
See all 168| Rank | Project | Score |
|---|---|---|
| 1 | transformersPython library defining open models for text, vision, audio and multimodal tasks | 94 out of 100 |
| 2 | pytorchPython tensor library with GPU support and tape-based autograd | 93 out of 100 |
| 3 | ultralyticsPython library and CLI for training and running YOLO vision models | 93 out of 100 |
| #4 | pytorch-image-modelsPyTorch library of image models, pretrained weights and training scripts | 93 out of 100 |
| #5 | supervisionPython toolkit for loading, annotating and counting computer vision detections | 92 out of 100 |