netron: a visual viewer for neural network and machine learning model architectures

What it solves

Netron is a visualizer for neural network and machine learning models. It allows users to see the architecture of a model by opening a model file, which is often a complex set of mathematical operations and layers. It simplifies the model inspection process by providing a graphical representation of the same.

How it works

Netron reads model files from various formats. It supports a wide range of weights and architecture files, including ONNX, TensorFlow Lite, PyTorch (TorchScript, torch.export, ExecuTorch), TensorFlow, Core ML, OpenVINO, Keras, Caffe, Darknet, and Safetensors, as well as NumPy arrays. It also has experimental support for MLIR, JAX, GGUF, RKNN, ncnn, MNN, PaddlePaddle, and scikit-learn.

Who it’s for

It is designed for developers and researchers who need to inspect, debug, or understand the same of a pre-trained model's structure and flow of data flow through the same.low

Highlights

• Comprehensive support for almost all major AI model formats.
• Available as a browser-based tool, desktop application (macOS, Linux, Windows), and a Python package.
• Ability to open sample models directly from the browser version for quick testing.