Empty canvas. Click New model for a starter network, or click any layer in the list on the left to add it.
Init varianceThe model changed since this plot.
PyTorch
Export for trainingmodel.py with forward(), loss() and predict(), and a train.py to run it
Import a model
Paste PyTorch module code (an nn.Module with __init__ and forward), JSON from export_graph.py, or graph JSON copied from this tool. To work on a file in place, use Open file instead, or drop the file on the canvas.
Models the code reader can't follow? Export them with torch.fx
The code reader handles nn.Linear, norms, activations, nn.Sequential, nn.ModuleList, loops and custom submodules. For anything else, run this next to your model and paste the JSON it writes. Other layer types come in as custom modules that keep their original constructor.
Torch Canvas