Model Explorer offers an intuitive and hierarchical visualization of model graphs. It organizes model operations into nested layers, enabling users to dynamically expand or collapse these layers. It also provides a range of features to facilitate model exploration and debugging, including the ability to highlight input and output operations, overlay metadata on nodes, display layers in interactive pop-ups, perform searches, show identical layers, GPU-accelerated graph rendering, among others. It currently supports TFLite, TF, TFJS, MLIR, and PyTorch (Exported Program) model format, and provides an extension framework for developers to easily add support for additional formats.
To start using Model Explorer, run:
$ pip install ai-edge-model-explorer
$ model-explorerPlease check out our Wiki for more details:
- Installation
- User Guide
- Command Line Guide
- API Guide
- Run in Colab Notebook
- Develop Adapter Extension
- Limitations and Known Issues
We invite you to participate in research studies on Model Explorer. Sign up here.
- ONNX Adapter: https://github.com/justinchuby/model-explorer-onnx
You are invited to create custom adapters to add support for additional model formats. Please refer to our Develop Adapter Extension guide. We will gladly consider pull requests to add links to your adapter's GitHub repository and PyPI package to this README.
- Introduction video on YouTube
- Blog post on Google Research Blog