torchstat

This is a lightweight neural network analyzer based on PyTorch. It is designed to make building your networks quick and easy, with the ability to debug them. Note: This repository is currently under development. Therefore, some APIs might be changed.

This tools can show

• Total number of network parameters
• Theoretical amount of floating point arithmetics (FLOPs)
• Theoretical amount of multiply-adds (MAdd)
• Memory usage

Installing

There're two ways to install torchstat into your environment.

• Install it via pip.

$ pip install torchstat

• Install and update using setup.py after cloning this repository.

$ python3 setup.py install

A Simple Example

If you want to run the torchstat asap, you can call it as a CLI tool if your network exists in a script. Otherwise you need to import torchstat as a module.

CLI tool

$ torchstat masato$ torchstat -f example.py -m Net
[MAdd]: Dropout2d is not supported!
[Flops]: Dropout2d is not supported!
[Memory]: Dropout2d is not supported!
      module name  input shape output shape     params memory(MB)           MAdd         Flops  MemRead(B)  MemWrite(B) duration[%]   MemR+W(B)
0           conv1    3 224 224   10 220 220      760.0       1.85   72,600,000.0  36,784,000.0    605152.0    1936000.0      57.49%   2541152.0
1           conv2   10 110 110   20 106 106     5020.0       0.86  112,360,000.0  56,404,720.0    504080.0     898880.0      26.62%   1402960.0
2      conv2_drop   20 106 106   20 106 106        0.0       0.86            0.0           0.0         0.0          0.0       4.09%         0.0
3             fc1        56180           50  2809050.0       0.00    5,617,950.0   2,809,000.0  11460920.0        200.0      11.58%  11461120.0
4             fc2           50           10      510.0       0.00          990.0         500.0      2240.0         40.0       0.22%      2280.0
total                                        2815340.0       3.56  190,578,940.0  95,998,220.0      2240.0         40.0     100.00%  15407512.0
===============================================================================================================================================
Total params: 2,815,340
-----------------------------------------------------------------------------------------------------------------------------------------------
Total memory: 3.56MB
Total MAdd: 190.58MMAdd
Total Flops: 96.0MFlops
Total MemR+W: 14.69MB

If you're not sure how to use a specific command, run the command with the -h or –help switches. You'll see usage information and a list of options you can use with the command.

Module

from torchstat import stat
import torchvision.models as models

model = models.resnet18()
stat(model, (3, 224, 224))

Features & TODO

Note: These features work only nn.Module. Modules in torch.nn.functional are not supported yet.

• FLOPs
• Number of Parameters
• Total memory
• Madd(FMA)
• MemRead
• MemWrite
• Model summary(detail, layer-wise)
• Export score table
• Arbitrary input shape

For the supported layers, check out the details.

Requirements

• Python 3.6+
• Pytorch 0.4.0+
• Pandas 0.23.4+
• NumPy 1.14.3+

References

Thanks to @sovrasov for the initial version of flops computation, @ceykmc for the backbone of scripts.

• flops-counter.pytorch
• pytorch_model_summary
• chainer_computational_cost
• convnet-burden.