PyTorch Network Loader

PyTorch Network Loader is a utility that enables the easy creation of neural networks in PyTorch using JSON configuration files. The package automatically tracks layer output shapes, eliminating the need to manually calculate input dimensions for each layer. NetLoader also provides an architecture framework that allows for the easy architecture setup and training framework, along with additional useful utilities to accelerate the use of neural networks in science.

See the Wiki for more information on how to use this package.
For a real-world example of this package, see Bayesian-DARKSKIES.

Requirements

Using Within Projects

• Install using pip install netloader¹
• Example of InceptionV4 can be downloaded under ./network_configs/inceptionv4.json along with the composite layers in ./network_configs/composite_layers/

Locally Running NetLoader

• Clone or download the repository
• Install dependencies: pip install -r requirements.txt
• PyTorch's dependencies²:
  NVIDIA GPU with CUDA Toolkit ~= v12.1

Footnotes

1. To use normalising flows or weights & biases, netloader must be pip installed with the optional argument flows and/or wandb: pip install netloader[flows,wandb] ↩

2. Only required for use with NVIDIA GPU, v11.8 is also supported, but requirements.txt will try to install the v12.1 version ↩