Neuromap SDK¶
Program neuromorphic chips with Python.
Neuromap is a Python SDK for building, training, quantizing, and
exporting spiking neural networks (SNNs) for Neuromap neuromorphic
hardware. It gives you a PyTorch-based workflow that goes from an
untrained model to a portable .nmap hardware bundle.
Why Neuromap¶
- Chip-aware networks. Define networks that match your target chip's topology, weight bit-width, and neuron parameters, so what you train is what you export.
- Batteries-included training. A single
Trainerclass handles LR warmup, scheduling, gradient clipping, early stopping, and best-model checkpointing. - Hardware export. Quantize weights to the chip's DAC bit-width and
package everything into a portable
.nmaparchive. - Streaming inference. Stateful chunk-by-chunk processing for real-time applications.
Install¶
pip install neuromap
The pipeline in one screen¶
from neuromap import Network, Trainer, Exporter, chips
# 1. Build a network matching the NeuroSoC-v1 chip
net = Network(chips.NEUROSOC_V1)
print(net.summary())
# 2. Train it (bring your own DataLoader)
trainer = Trainer(net, lr=1e-3, epochs=20)
history = trainer.fit(train_loader, val_loader)
# 3. Export a hardware bundle
Exporter(net).quantize().save("model.nmap")
Where to go next¶
- Quickstart walks through the same pipeline step by step, with runnable snippets.
- Concepts explains the four-stage model behind the SDK: chip spec, network, trainer, and exporter.
- API Reference is the full class-by-class reference, generated from the source docstrings.