lutzroeder/netron is a visualization tool for laying neural network, deep learning and machine learning models out as graphs. The repository description defines the tool as a model visualizer, states that it is offered as a browser-based web version, a desktop app and a Python package, and that sample models can be opened. The latest release in the repository is v9.3.2.
Opening a model reveals the architecture and operations, plus weight metadata, as graphs and tensors. Supported formats are ONNX, TensorFlow Lite, PyTorch, TensorFlow, Core ML and OpenVINO, while MLIR, JAX and GGUF are at the experimental stage. Its place is that you can go through several model files of different formats the same way in a single tool. The area it handles directly, within model analysis, is checking architecture, operations and weight metadata. The code is written in JavaScript and the license is MIT. It has 33596 GitHub stars, and the repository was created on December 26, 2010.
That said, the record compiled this time contains no release notes or changelog for v9.3.2. What this version fixed cannot be known, and that does not mean there are no public improvements. There are also no performance comparisons or processing-speed figures, so there is not enough basis to measure the actual size of the improvement. MLIR, JAX and GGUF are experimental, so these three formats can be considered less mature, and how far the analysis flow has changed will not be judged until the release record is checked.