[Ecosystem] VisualTorch

[willyfh]

Contact emails

willyfitrahendria@gmail.com

Project summary

A PyTorch model visualization library for generating publication-ready diagrams of neural network architectures.

Project description

VisualTorch is a visualization toolkit designed for PyTorch users to generate clean, publication-ready diagrams of neural network architectures with a single function call. It supports multiple visualization styles, layered, graph-style, and LeNet-style, and works seamlessly with both nn.Sequential and custom nn.Module models.

Given a model and input shape, VisualTorch traces a forward pass using dummy input and automatically renders an architectural diagram. The output can be customized with options such as color, opacity, sizing, and legends, making it suitable for debugging, teaching, and scientific publications.

The library fills a long-standing gap in the PyTorch ecosystem by providing an inline, framework-native visualization tool.

The project is open-source under the MIT license and supported by web-based documentation, and CI/CD workflows. VisualTorch has also been published in the Journal of Open Source Software (JOSS).

Are there any other projects in the PyTorch Ecosystem similar to yours? If, yes, what are they?

Similar projects include:

• VisualKeras: a visualization tool for Keras models that inspired VisualTorch.

VisualTorch is distinct in its tight integration with PyTorch, its ease of use within Python scripts and notebooks. It allows rapid, high-quality visualization with minimal user effort, making it ideal for researchers and engineers working in PyTorch.

Project repo URL

https://github.com/willyfh/visualtorch

Additional repos in scope of the application

No response

Project license

MIT

GitHub handles of the project maintainer(s)

willyfh

Is there a corporate or academic entity backing this project? If so, please provide the name and URL of the entity.

No official backing. Developed by an independent researcher.

Website URL

https://visualtorch.readthedocs.io/en/latest/

Documentation

Yes. The documentation is hosted on ReadTheDocs:

https://visualtorch.readthedocs.io/en/latest/

It includes:

• Getting started and installation
• Multiple usage examples (layered, graph, LeNet)
• Full API reference
• Developer and contribution guides
• Citation metadata

How do you build and test the project today (continuous integration)? Please describe.

VisualTorch uses GitHub Actions for continuous integration and release automation. Every pull request and push to the main branch triggers automated workflows that include:

• Pre-commit checks using pre-commit and ruff for code formatting, linting, and static analysis
• Static type checking with mypy to ensure type safety across the codebase
• Unit testing using pytest to validate functionality across supported model types and visualization styles
• Dependency management via optional development extras (pip install -e .[dev]), ensuring reproducibility across environments

Contributors follow a documented setup process that includes environment creation, pre-commit installation, and test verification. CI ensures that every contribution adheres to the project’s quality standards before merging.

In addition, VisualTorch uses GitHub Actions for automated release publishing to PyPI, making new versions available as soon as they are tagged and validated—streamlining distribution and ensuring users always have access to the latest stable version.

Version of PyTorch

Compatible and tested with PyTorch ≥ 2.0+

Components of PyTorch

• torch.nn.Module: for model parsing and layer introspection
• torch.Tensor: for shape tracing during the forward pass

How long do you expect to maintain the project?

Committed to maintaining the project for at least 10 more years, including updates for new PyTorch releases and ongoing feature enhancements. Actively maintained on GitHub with CI pipelines and public issue tracking.

Additional information

VisualTorch has been peer-reviewed and published in the Journal of Open Source Software (JOSS):

Citation:
Willy Fitra Hendria & Paul Gavrikov. (2024). VisualTorch: Streamlining Visualization for PyTorch Neural Network Architectures. JOSS. DOI

The library has gained adoption among researchers needing quick, publication-ready model diagrams. It is designed to be framework-native, beginner-friendly, and production-agnostic—ideal for debugging, teaching, and scientific communication.

PyPI: https://pypi.org/project/visualtorch
Install via: pip install visualtorch

[atsyplikhin]

Hi Willy & team — thank you for submitting VisualTorch! We appreciate the nice docs, CI, tests, PyPI packaging, et al. — this project fills a useful gap.

To meet the PyTorch Ecosystem requirements (link), please consider addressing the following:

• ("must") Core maintainers (requirement = 2): Currently 1 listed (willyfh). Please add at least one additional core maintainer and list them in your governance file.
• ("should") Contributors (requirement ≥ 5 in last 90 days): Currently 2 contributors active. Encourage contributions (labels like good first issue / help wanted, outreach, or invite co-maintainers) to reach ≥5 active contributors.
• ("must") Commits (requirement ≥ 30 in last 90 days): There are ~3 commits in the recent 90-day window (Jun 23, Jul 3, Jul 6, 2025). Increasing cadence (or documenting ongoing maintenance plans in GOVERNANCE.md) might help.
• ("must") Stars (requirement ≥ 200): Current stars ≈ 94. Community outreach, sharing your JOSS paper & examples, or including the project in tutorials can help grow visibility.
• ("must") Governance: Add a GOVERNANCE.md (or similar) describing decision-making, maintainer roles and process.
• CODEOWNERS: Add a CODEOWNERS file to formally document code ownership and reviewers.
• License file: Rename the root license to LICENSE.md for consistency.

You’re very close — the technical requirements (tests, docs, CI, PyPI package, permissive license, quick start) look solid. Once you add the governance/CODEOWNERS files and make progress on the contributor/commit/star targets, we’ll be happy to re-review for ecosystem approval.

Thanks again — great work so far!