neuromap 0.1.4

SDK for programming Neuromap neuromorphic chips

Neuromap SDK

Python SDK for programming Neuromap neuromorphic chips.

Installation

pip install neuromap

For development (using Poetry):

cd neuro-sim
poetry install --with dev

Quick Start

from neuromap import Network, Trainer, Exporter, chips

# Build a network matching the NeuroSoC-v1 chip
net = Network(chips.NEUROSOC_V1)
print(net.summary())

# Train
trainer = Trainer(net, lr=1e-3, epochs=20)
history = trainer.fit(train_loader, val_loader)

# Export for deployment
Exporter(net).quantize().save("model.nmap")

Simplified API (snnTorch)

The SDK provides convenience functions backed by snnTorch for a streamlined workflow:

import neuromap as nm

# Encode raw data to spike trains (wraps snntorch.spikegen)
spike_data = nm.encode_sensor_data(raw_images, num_steps=100)

# Build network from the fixed chip topology
model = nm.Network(nm.chips.NEUROSOC_V1)

# Configure surrogate gradients for training
nm.compile_model(model, surrogate="atan")

# Train, then quantize & export to hardware
nm.quantize_and_export_to_spi(model, path="model.nmap")

Available surrogate gradients: "atan" (default), "fast_sigmoid", "straight_through", "spike_rate_escape".

Available encoding methods: "rate" (default), "latency", "delta".

Running Tests

poetry run pytest tests/ -v

Or via Nx from the repo root:

npx nx run neuro-sim:test

Linting & Formatting

poetry run ruff check .
poetry run ruff format .