Synaptic OPerations (SyOPs) counter for spiking neural networks

This script is designed to compute the theoretical amount of synaptic operations in spiking neural networks, including accumulated (AC) and multiply-accumulate (MAC) operations. It can also compute the number of parameters and print per-layer computational cost of a given network. This tool is still under construction. Comments, issues, contributions, and collaborations are all welcomed!

Supported layers:

• Conv1d/2d/3d (including grouping)
• ConvTranspose1d/2d/3d (including grouping)
• BatchNorm1d/2d/3d, GroupNorm, InstanceNorm1d/2d/3d
• Activations (ReLU, PReLU, ELU, ReLU6, LeakyReLU, GELU)
• Linear
• Upsample
• Poolings (AvgPool1d/2d/3d, MaxPool1d/2d/3d and adaptive ones)
• LF/LIF/PLIF (spikingjelly)

Experimental support:

• RNN, LSTM, GRU (NLH layout is assumed)
• RNNCell, LSTMCell, GRUCell
• MultiheadAttention

Requirements: Pytorch >= 1.1, torchvision >= 0.3, spikingjelly<=0.0.0.0.12

Usage

• This script doesn't take into account torch.nn.functional.* operations. For an instance, if one have a semantic segmentation model and use torch.nn.functional.interpolate to upscale features, these operations won't contribute to overall amount of flops. To avoid that one can use torch.nn.Upsample instead of torch.nn.functional.interpolate.
• syops launches a given model on a random tensor or a DataLoader and estimates amount of computations during inference. Complicated models can have several inputs, some of them could be optional.
  • To construct non-trivial input one can use the input_constructor argument of the get_model_complexity_info. input_constructor is a function that takes the input spatial resolution as a tuple and returns a dict with named input arguments of the model. Next this dict would be passed to the model as a keyword arguments.
  • To construct a DataLoader input one can use the dataLoader argument of the get_model_complexity_info based on torch.utils.data.DataLoader. The number of computations would be estimated based on the input fire rate of spike signals.
• verbose parameter allows to get information about modules that don't contribute to the final numbers.
• ignore_modules option forces syops to ignore the listed modules. This can be useful for research purposes. For an instance, one can drop all batch normalization from the counting process specifying ignore_modules=[torch.nn.BatchNorm2d].

Install the latest version

From PyPI:

pip install syops

From this repository:

pip install --upgrade git+https://github.com/iCGY96/syops-counter

Example

import torch
from spikingjelly.activation_based import surrogate, neuron, functional
from spikingjelly.activation_based.model import spiking_resnet
from syops import get_model_complexity_info

dataloader = ...
with torch.cuda.device(0):
    net = spiking_resnet.spiking_resnet18(pretrained=True, spiking_neuron=neuron.IFNode, 
            surrogate_function=surrogate.ATan(), detach_reset=True)
    ops, params = get_model_complexity_info(net, (3, 224, 224), dataloader, as_strings=True,
                                            print_per_layer_stat=True, verbose=True)
    print('{:<30}  {:<8}'.format('Computational complexity ACs:', acs))
    print('{:<30}  {:<8}'.format('Computational complexity MACs:', macs))
    print('{:<30}  {:<8}'.format('Number of parameters: ', params))

Benchmark

• ACs(G) - The theoretical amount of accumulated operations based on spike signals.
• MACs(G) - The theoretical amount of multiply-accumulate operations based on non-spike signals.
• Energy(mJ) - Energy consumption is based on 45nm technology, where AC cost 0.9pJ and MAC cost 4.6pJ.
• Acc@1 - ImageNet single-crop top-1 accuracy on validation images of the same size used during the training process.
• Acc@5 - ImageNet single-crop top-5 accuracy on validation images of the same size used during the training process.

Citation

If you find our work useful for your repo, please consider giving a star :star: and citation :beer::

@article{chen2023training,
  title={Training Full Spike Neural Networks via Auxiliary Accumulation Pathway},
  author={Chen, Guangyao and Peng, Peixi and Li, Guoqi and Tian, Yonghong},
  journal={arXiv preprint arXiv:2301.11929},
  year={2023}
}

Acknowledgements

This repository is developed based on ptflops