Skip to content
Get packs

Guide · Open source

How to use Netron: open and view neural network models

Updated · written with AI from the project's own documentation and checked against our catalogue data

Netron is a viewer for neural network, deep learning and machine learning models. You open a trained model file and it draws the layers and connections as an interactive diagram, so you can inspect a model's structure without writing code. This guide shows how to start Netron in the browser, on your desktop or from Python, and what to check once a model is open.

Netronlutzroeder/netron · ★ 33.6k · MIT

Quick start

  1. 1Open the browser version. Go to https://netron.app to start Netron without installing anything.
  2. 2Pick an install method. Choose the browser version, a desktop download for macOS, Linux or Windows, or the Python package.
  3. 3Install the Python package if needed. Run pip install netron in your terminal.
  4. 4Open your model file. Load your model in the browser, or run netron followed by the file path.
  5. 5Explore the layer diagram. Follow the layers and connections from input to output to check the architecture.
  6. 6Confirm the format is supported. Compare your file type with the formats listed in the README.

Open Netron in the browser without installing anything

The quickest way to use Netron is the browser version at netron.app. The README lists it as the install method for the browser app and names no prerequisites.

The README also links sample model files that you can open directly in the browser version, including ONNX squeezenet, TensorFlow Lite yamnet and Keras mobilenet.

  • Go to https://netron.app
  • Open one of the sample links from the README, or load your own model file
  • Explore the diagram that appears

Install Netron on macOS, Linux or Windows

The README gives a separate option for each desktop platform. On macOS you can download the .dmg file or run a Homebrew command. On Linux you can download a .deb or .rpm file. On Windows you can download the .exe installer or use winget. All downloads are linked from the project's latest releases page.

The README does not describe any other installation requirements.

  • macOS: download the .dmg file, or run the Homebrew command below
  • Linux: download the .deb or .rpm file
  • Windows: download the .exe installer, or run the winget command below
brew install --cask netron winget install -s winget netron

Open a model from Python with pip

If you work in Python, install the package with pip. Then you can open a model file from the command line by passing its path to netron, or call netron.start from your own script.

Replace [FILE] with the path to your model file.

  • Install the package with pip
  • Run netron followed by the path to your model file
  • Or call netron.start with the file path inside a Python script
pip install netron netron [FILE]

Read the diagram of a model

Once a model is open, Netron shows its layers as nodes and the connections between them as edges, so you can follow data from input to output. The README describes the tool as showing a model's architecture, which makes it useful for checking layer order and tensor connections before you deploy a model.

The README does not document detailed controls for the diagram, so use the on-screen options to explore it.

  • Follow the connections from the input layer to the output layer
  • Check that the layer order matches what you expect
  • Use the diagram to confirm layers after converting a model to another format

Check that your model format is supported

Netron supports ONNX, TensorFlow Lite, PyTorch, torch.export, ExecuTorch, TorchScript, TensorFlow, Core ML, OpenVINO, Keras, Caffe, Darknet, Safetensors and NumPy. It has experimental support for MLIR, JAX, GGUF, RKNN, ncnn, MNN, PaddlePaddle and scikit-learn.

If your format is not in these lists, the README does not say Netron will open it.

  • Fully listed: ONNX, TensorFlow Lite, PyTorch, torch.export, ExecuTorch, TorchScript, TensorFlow, Core ML, OpenVINO, Keras, Caffe, Darknet, Safetensors, NumPy
  • Experimental: MLIR, JAX, GGUF, RKNN, ncnn, MNN, PaddlePaddle, scikit-learn

Know what Netron does not do

Netron is a viewer. The README describes it as a viewer, not an editor, so it is not for training or changing models. The README also does not describe debugging or performance profiling features, so do not expect them.

  • Does not train models
  • Does not edit models (the README describes it as a viewer)
  • Does not describe debugging or performance profiling features

What people use it for

  • Checking the layers and tensor connections of a model before deploying it
  • Learning how a neural network is structured by viewing a real model file
  • Confirming that a model converted to another format has the expected structure
  • Viewing a sample model quickly in the browser without installing software

Frequently asked questions

Is Netron free to use?

Netron is open source under the MIT licence, as listed on the project page. The README does not mention any paid tier.

Do I need to install anything to use Netron?

No. The README lists a browser version at netron.app, and the README does not list any prerequisites. Desktop downloads and the Python package are also available if you prefer them.

How do I open a model file in Netron from Python?

Install the package with pip install netron, then run netron followed by your file path. You can also call netron.start with the file path inside a script.

Which model formats does Netron support?

The README lists ONNX, TensorFlow Lite, PyTorch, torch.export, ExecuTorch, TorchScript, TensorFlow, Core ML, OpenVINO, Keras, Caffe, Darknet, Safetensors and NumPy. It also lists experimental support for MLIR, JAX, GGUF, RKNN, ncnn, MNN, PaddlePaddle and scikit-learn.

Can Netron train or edit a model?

No. The README describes Netron as a viewer for neural network, deep learning and machine learning models. It is for inspecting architecture, not training or editing.

What are alternatives to Netron?

Comparable projects named in the research are Netscope, Keras Visualizer and TensorBoard. Choose based on your model format and whether you need a viewer only.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Netron suits engineers checking a model's layers before deployment, students studying a real network file, and developers confirming a format conversion. If you need to train or edit models, or a full debugger or profiler, pick a different tool, since the README describes Netron only as a viewer.

Install options & downloads