ONNX Model Comparator for VS Code

Compare two ONNX (Open Neural Network Exchange) models side-by-side within Visual Studio Code and visualize their differences.

You can perceive this as a beyond compare for ONNX models.

Overview

Features

Select Models: Easily select two .onnx model files from your workspace using file pickers.
Side-by-Side View: Displays comparison results in a dedicated webview panel.
Compare Key Components:
- Model Information: Compares metadata like IR version, producer, model version.
- Inputs: Shows differences in input names, data types, and shapes.
- Outputs: Shows differences in output names, data types, and shapes.
- Nodes (Operators): Lists added, removed, or modified nodes, detailing changes in inputs, outputs, attributes, and parameter shapes.
- Initializers: Compares weights and constants, highlighting differences in data type and shape.
Highlight Differences: Uses color-coding (green for added, red for removed, orange for modified) to easily spot changes.
Node Search: Filter the nodes table by name to quickly find specific operators.

Open the VS Code Command Palette (Cmd+Shift+P on macOS, Ctrl+Shift+P on Windows/Linux).
Run the command: Compare ONNX Models.
You will be prompted to select the first ONNX model file.
You will then be prompted to select the second ONNX model file.
A new panel will open displaying the comparison results.

Comparison of large initializers (weights) currently only checks for existence, data type, and shape differences, not value differences.
Display of very complex tensor shapes or attributes might be simplified.

This extension is licensed under the MIT License.