Skip to content

zhijian-liu/torchprofile

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Torchprofile

This is a profiler to count the number of MACs / FLOPs of PyTorch models based on torch.jit.trace.

  • It is more general than ONNX-based profilers as some operations in PyTorch are not supported by ONNX for now.
  • It is more accurate than hook-based profilers as they cannot profile operations within torch.nn.Module.

Installation

pip install torchprofile

Getting Started

You should first define your PyTorch model and its (dummy) input:

import torch
from torchvision.models import resnet18

model = resnet18()
inputs = torch.randn(1, 3, 224, 224)

You can then measure the number of MACs using profile_macs:

from torchprofile import profile_macs

macs = profile_macs(model, inputs)

License

This repository is released under the MIT license. See LICENSE for additional details.

About

A general and accurate MACs / FLOPs profiler for PyTorch models

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages