Open-source project
btorch
Brain-inspired differentiable PyTorch toolkit for neuromorphic and computational neuroscience research.
The project
Build and explore neural circuits.
BTorch is our lab's flagship open-source Python framework, evolved from SpikingJelly and influenced by BrainState, for building, simulating, and training recurrent spiking neural networks and biologically grounded circuit models directly in native PyTorch.
BTorch provides a unified software substrate for our NeuroAI program, connecting sparse connectome-style architectures and heterogeneous neuron-synapse dynamics to scalable gradient-based and local learning.
- Recurrent spiking neural network modeling with heterogeneous cellular parameters and stateful neuron and synapse modules using explicit buffer-managed memory.
- Support for sparse and connectome-style network topologies alongside integrated analysis tools for dynamical criticality and Lyapunov spectra.
- Native PyTorch training integration including torch.compile compatibility, gradient checkpointing, truncated backpropagation through time, and ONNX export.
Get started
Install btorch
Install the latest released package from PyPI:
pip install btorchExplore the documentation and examples for details.