Ctrl+K

Sinabs

Sinabs#

Sinabs is a deep learning library for spiking neural networks which is based on PyTorch and focuses on fast training as well as inference on neuromorphic hardware.

sinabs.to_nir and sinabs.from_nir methods allow you to seemlessly navigate between nir and sinabs. Once your model is in sinabs, you can use this model to train or directly deploy your models to Speck/DynapCNN.

Import a NIR graph to Sinabs#

import torch
from sinabs import from_nir

import nir

# Create a NIR graph
affine_weights = torch.tensor([[1.0, 2.0], [3.0, 4.0]])
affine_bias = torch.tensor([1.0, 2.0])
li_tau = torch.tensor([0.9, 0.8])
li_r = torch.tensor([1.0, 1.0])
li_v_leak = torch.tensor([0.0, 0.0])
nir_network = nir.NIRGraph.from_list(
    nir.Affine(affine_weights, affine_bias), nir.LI(li_tau, li_r, li_v_leak)
)

# Create Sinabs model from NIR graph.
# You need to define the batch size because Sinabs will use Squeeze
# versions of layers by default.
sinabs_model = from_nir(nir_network, batch_size=10)
print(sinabs_model)

Export a NIR graph from Sinabs#

import sinabs.layers as sl
import torch
import torch.nn as nn
from sinabs import from_nir, to_nir

batch_size = 4

# Create Sinabs model
orig_model = nn.Sequential(
    torch.nn.Linear(10, 2),
    sl.ExpLeakSqueeze(tau_mem=10.0, batch_size=batch_size),
    sl.LIFSqueeze(tau_mem=10.0, batch_size=batch_size),
    torch.nn.Linear(2, 1),
)

# Convert model to NIR graph with a random input of representative shape
nir_graph = to_nir(orig_model, torch.randn(batch_size, 10))
print(nir_graph)


# Reload sinabs model from NIR
sinabs_model = from_nir(nir_graph, batch_size)
Contents