snntorch

Why we included this project

For anyone building spiking neural networks, snnTorch keeps you inside the PyTorch ecosystem you already know. It models each spiking neuron as a recurrent activation unit within a standard torch.nn-style model, so you can keep using familiar layers, autograd, and CUDA acceleration while exploring spike-based computation. The package includes pre-designed neuron models, surrogate gradient functions for training, spike generation and conversion utilities, plus visualization and dataset helpers, which handles much of the boilerplate that usually slows down SNN research. It also supports importing and exporting models through the NIR interchange format, making it practical to move networks between different spiking frameworks. Researchers, students, and engineers working on neuromorphic or energy-efficient computing will find the tutorials and Colab notebooks a good starting point, and the library runs small training jobs on CPU when needed.