📊 Supported Models¶

NNViz supports model loading from the following sources:

• Torchvision built-in models (default)

• Models from any installed Python package

• Models from any local Python file

Visualizing built-in torchvision models¶

By default, NNViz will attempt to load models from the torchvision.models module. You can pass the name of any model contained in torchvision.models and NNViz will load it for you.

These are all valid examples:

nnviz resnet18
nnviz efficientnet_b0
nnviz convnext_tiny

Visualizing models from installed packages¶

Being limited to torchvision models is not very useful, so you can also visualize models from any installed package, as long as they are importable. To do so, you need to pass the fully qualified name of a python function or class that returns an instance of the model you want to visualize.

Syntax: <module>:<symbol>

NNViz only cares that a package with the given name exists in the Python path, and that the symbol is either:

• A nn.Module instance

• A function that returns a nn.Module instance

• A nn.Module subclass

Examples:

nnviz package.module.submodule:my_net_instance
nnviz package.module.submodule:MyModelClass
nnviz package.module.submodule:a_function_that_returns_a_model

Visualizing models from python files¶

And if want to visualize a model that is not in a package, you can also pass a path to a python file that contains the model constructor. NNViz will attempt to dynamically import the file and extract the model constructor from it. The same rules apply as for installed packages, the only difference is that you need to pass the path to the file instead of the package name.

Syntax: <path>:<symbol>

nnviz path/to/file.py:my_net_instance
nnviz path/to/file.py:MyModelClass
nnviz path/to/file.py:a_function_that_returns_a_model

Passing arguments¶

You can also pass simple built-in type arguments to the function, by adding ; separated key-value pairs after the function name. Arguments are parsed with eval, be careful with what you pass.

Syntax: <model>;<key1>=<value1>;<key2>=<value2>...

Examples:

nnviz package.module.submodule:my_net_instance;num_classes=10;dropout=0.2
nnviz package.module.submodule:MymodelClass;width=224
nnviz package.module.submodule:a_function_that_returns_a_model;num_classes=10;dropout=0.2

nnviz path/to/file.py:my_net_instance;num_classes=10;dropout=0.2
nnviz path/to/file.py:MyModelClass;width=224
nnviz path/to/file.py:a_function_that_returns_a_model;num_classes=10;dropout=0.2