spikeforge

A toolkit for building, training, and deploying spiking neural networks (SNNs), built on snnTorch and PyTorch.

Load MNIST-style or neuromorphic event datasets, encode them into spikes, train LIF networks, and check what happens when you export or deploy them. Use it as a Python library, from the command line, through a browser dashboard, or as a desktop app.

Pre-1.0. Before trusting any number this produces, read Implications and boundaries.

Quickstart

Install

Install from pypi

pip install spikeforge

Then run the following snippet in a Python REPL or script:

from spikeforge import TrainingEngine

engine = TrainingEngine(dataset="mnist", hidden=32, epochs=1, num_steps=5)
for metrics in engine.train():
    last = metrics
print(last)  # {'loss': ..., 'train_accuracy': ..., 'test_accuracy': ...}

That finishes in seconds: about 3 s on a 12-thread desktop CPU, 14 s on a laptop. Accuracy lands in the mid-80s.

The snippet sets no seed, so the exact number moves between runs. The test_accuracy it prints is a fast progress probe, not the whole test split. For numbers on the complete held-out split, with the command that reproduces each one, see Benchmarks.

The first run downloads MNIST. Later runs are offline.

TrainingEngine is the entry point for new code. It owns the training loop, topology selection, encoding, checkpointing, and evaluation.

SNNTrainer is also exported, but it's the older MNIST rate-coding helper behind the tutorial demo. Use TrainingEngine for anything new.

The CLI tools (spikeforge-verify, spikeforge-benchmark) and the NIR/deployment/energy pieces are covered in Usage.

No GPU

Install the CPU torch wheels first, so pip doesn't pull in the full CUDA stack behind them:

pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
pip install spikeforge

That's a ~1.1 GB environment instead of ~5.5 GB. It's the same flow the Dockerfile and CI use.

The order matters: with --extra-index-url pip still prefers the CUDA build from PyPI.

Browser dashboard

To run the live browser dashboard locally with Docker:

git clone https://github.com/Capsize-Games/spikeforge.git
cd spikeforge
docker compose up --build

Open http://localhost:8877. The dashboard connects to the WebSocket on the same host and port — no separate backend or proxy to run.

The desktop app runs the same dashboard without Docker.

See Usage and Quickstart for every path (./install.sh, local dev with Vite, examples/, and the spikeforge-* console scripts).

Why not just snnTorch, Norse, Lava, or SpikingJelly?

spikeforge doesn't replace snnTorch, Norse, or Lava. It trains through snnTorch and can deploy to Norse and Lava.

Here's what it adds on top:

• An independent NIR interpreter that re-executes every exported graph and reports numerical drift against the original model. (snnTorch and Norse can also export NIR; Lava-DL only reads it, not writes it.)
• A deployment-capability matrix that reports per-target op support and honest availability, not just a support claim.
• A built-in event-driven energy estimator.
• A live training/introspection dashboard.

To our knowledge, none of the other libraries ship the last two as part of the library itself — file an issue if that's stale.

If you already have a training loop in snnTorch, Norse, Lava, or SpikingJelly, and you don't need interop, deployment reporting, energy estimates, or the dashboard, you may not need spikeforge on top of it.

Features

What's implemented vs. experimental vs. spec-only

The features list above spans very different levels of maturity. This is the ten-second version; each row links to the honest detail.

What it scores

Reference configurations on the datasets and topologies that ship, each measured on the complete held-out test split, with the command that reproduces it:

Every figure is the complete held-out test split, on a 12-thread x86-64 CPU with no GPU.

The full table adds epochs, time steps, hardware, seed, and the exact command that reproduces each row. See Benchmarks.

These are reference configurations with stock hyperparameters and a single seed — not tuned attempts at state of the art. Read them as a floor the shipped defaults reach, not a ceiling.

The checkpoints they produce are the trained entries in the model hub (spikeforge-hub list --trained).

Packages

This repository is a single workspace that publishes seven distributions, each versioned independently:

The satellites sit at lower version numbers than core by design, not neglect. Each package is versioned independently and only moves when it changes.

compatibility.json is the source of truth for which satellite versions go with which core release. If you're pinning versions by hand, read that file — don't assume semver alignment across packages.

Documentation

This README covers first contact and positioning only. For more:

• documentation/ — the full reference: install paths, CLI tools, architecture, module layout, and the dev workflow. Written for contributors and coding agents alike.
• COOKBOOK.md — copy-pasteable recipes.
• examples/ — runnable end-to-end scripts.
• plans/ — design documents and the roadmap.

See CONTRIBUTING.md and rules.md before opening a pull request.

Citing

If spikeforge is useful in your research, please cite it — see CITATION.cff (GitHub renders a "Cite this repository" button from it automatically).

License

Released under the BSD 3-Clause License — see LICENSE and AUTHORS.