https://github.com/nemonameless/pytorch_summary
Calculate parameters quantity, memory, and flops in simple and complex way.
https://github.com/nemonameless/pytorch_summary
Last synced: about 1 year ago
JSON representation
Calculate parameters quantity, memory, and flops in simple and complex way.
- Host: GitHub
- URL: https://github.com/nemonameless/pytorch_summary
- Owner: nemonameless
- License: mit
- Created: 2019-04-02T17:35:20.000Z (about 7 years ago)
- Default Branch: master
- Last Pushed: 2019-03-20T14:04:45.000Z (over 7 years ago)
- Last Synced: 2025-02-17T06:30:26.888Z (over 1 year ago)
- Language: Python
- Homepage:
- Size: 28.3 KB
- Stars: 2
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# PyTorch_summary
PyTorch summary:
Calculate parameters quantity, memory, run-time, and flops in simple and complex way.
# NEW
- Support run-time test.
- Support both GPU and CPU for **simple mode** and **complex mode**
- PyTorch 1.0 and 0.4 supported.
## Useage:
Add your model to summary.py
```
###### add your model ######
from torchvision import models
model = models.resnet50()
############################
```
Run summary.py
```
python3 summary.py --mod simple --gpu -1 --size 3,224,224
--mod simple|complex|val
--gpu (gpu id, -1:use cpu)
--size (input size, split with ,)
--runtime (if runtime is -1: disable runtime>=1 iters for runtime test.)
```
## Results:
### simplesum mode:
```
+ Number of params: 25.56M
+ Number of FLOPs: 4.11G
```
### run-time test:
```
Testing runtime using TITAN X (Pascal)
Run-Time: 141.8995 ms
```
### complexsum mode:
following https://github.com/ceykmc/pytorch_model_summary , and add the support to bilinear layer.
```
module name input shape output shape parameter quantity inference memory(MB) MAdd duration percent
0 conv1_Conv2d 3 224 224 64 112 112 9408 3.06MB 235,225,088 6.32%
1 bn1_BatchNorm2d 64 112 112 64 112 112 128 3.06MB 3,211,264 0.91%
2 relu_ReLU 64 112 112 64 112 112 0 3.06MB 802,816 0.94%
3 maxpool_MaxPool2d 64 112 112 64 56 56 0 0.77MB 1,605,632 3.58%
4 layer1.0.conv1_Conv2d 64 56 56 64 56 56 4096 0.77MB 25,489,408 0.45%
5 layer1.0.bn1_BatchNorm2d 64 56 56 64 56 56 128 0.77MB 802,816 0.21%
6 layer1.0.conv2_Conv2d 64 56 56 64 56 56 36864 0.77MB 231,010,304 2.75%
7 layer1.0.bn2_BatchNorm2d 64 56 56 64 56 56 128 0.77MB 802,816 0.22%
8 layer1.0.conv3_Conv2d 64 56 56 256 56 56 16384 3.06MB 101,957,632 1.38%
9 layer1.0.bn3_BatchNorm2d 256 56 56 256 56 56 512 3.06MB 3,211,264 0.78%
10 layer1.0.relu_ReLU 256 56 56 256 56 56 0 3.06MB 802,816 0.53%
11 layer1.0.downsample 64 56 56 256 56 56 16896 6.12MB 105,168,896 2.13%
12 layer1.1 256 56 56 256 56 56 70400 12.25MB 441,147,392 6.28%
13 layer1.2 256 56 56 256 56 56 70400 12.25MB 441,147,392 6.24%
14 layer2.0.conv1_Conv2d 256 56 56 128 56 56 32768 1.53MB 205,119,488 1.94%
15 layer2.0.bn1_BatchNorm2d 128 56 56 128 56 56 256 1.53MB 1,605,632 0.20%
16 layer2.0.conv2_Conv2d 128 56 56 128 28 28 147456 0.38MB 231,110,656 2.65%
17 layer2.0.bn2_BatchNorm2d 128 28 28 128 28 28 256 0.38MB 401,408 0.08%
18 layer2.0.conv3_Conv2d 128 28 28 512 28 28 65536 1.53MB 102,359,040 0.96%
19 layer2.0.bn3_BatchNorm2d 512 28 28 512 28 28 1024 1.53MB 1,605,632 0.25%
20 layer2.0.relu_ReLU 512 28 28 512 28 28 0 1.53MB 401,408 0.26%
21 layer2.0.downsample 256 56 56 512 28 28 132096 3.06MB 206,725,120 1.93%
22 layer2.1 512 28 28 512 28 28 280064 6.12MB 438,939,648 4.36%
23 layer2.2 512 28 28 512 28 28 280064 6.12MB 438,939,648 4.47%
24 layer2.3 512 28 28 512 28 28 280064 6.12MB 438,939,648 4.44%
25 layer3.0.conv1_Conv2d 512 28 28 256 28 28 131072 0.77MB 205,320,192 1.48%
26 layer3.0.bn1_BatchNorm2d 256 28 28 256 28 28 512 0.77MB 802,816 0.14%
27 layer3.0.conv2_Conv2d 256 28 28 256 14 14 589824 0.19MB 231,160,832 2.02%
28 layer3.0.bn2_BatchNorm2d 256 14 14 256 14 14 512 0.19MB 200,704 0.08%
29 layer3.0.conv3_Conv2d 256 14 14 1024 14 14 262144 0.77MB 102,559,744 0.77%
30 layer3.0.bn3_BatchNorm2d 1024 14 14 1024 14 14 2048 0.77MB 802,816 0.28%
31 layer3.0.relu_ReLU 1024 14 14 1024 14 14 0 0.77MB 200,704 0.15%
32 layer3.0.downsample 512 28 28 1024 14 14 526336 1.53MB 206,123,008 1.76%
33 layer3.1 1024 14 14 1024 14 14 1117184 3.06MB 437,835,776 3.97%
34 layer3.2 1024 14 14 1024 14 14 1117184 3.06MB 437,835,776 4.06%
35 layer3.3 1024 14 14 1024 14 14 1117184 3.06MB 437,835,776 4.05%
36 layer3.4 1024 14 14 1024 14 14 1117184 3.06MB 437,835,776 4.03%
37 layer3.5 1024 14 14 1024 14 14 1117184 3.06MB 437,835,776 4.02%
38 layer4.0.conv1_Conv2d 1024 14 14 512 14 14 524288 0.38MB 205,420,544 1.42%
39 layer4.0.bn1_BatchNorm2d 512 14 14 512 14 14 1024 0.38MB 401,408 0.14%
40 layer4.0.conv2_Conv2d 512 14 14 512 7 7 2359296 0.10MB 231,185,920 2.41%
41 layer4.0.bn2_BatchNorm2d 512 7 7 512 7 7 1024 0.10MB 100,352 0.13%
42 layer4.0.conv3_Conv2d 512 7 7 2048 7 7 1048576 0.38MB 102,660,096 1.03%
43 layer4.0.bn3_BatchNorm2d 2048 7 7 2048 7 7 4096 0.38MB 401,408 0.44%
44 layer4.0.relu_ReLU 2048 7 7 2048 7 7 0 0.38MB 100,352 0.08%
45 layer4.0.downsample 1024 14 14 2048 7 7 2101248 0.77MB 205,821,952 2.48%
46 layer4.1 2048 7 7 2048 7 7 4462592 1.53MB 437,283,840 5.26%
47 layer4.2 2048 7 7 2048 7 7 4462592 1.53MB 437,283,840 5.27%
48 avgpool_AvgPool2d 2048 7 7 2048 1 1 0 0.01MB 100,352 0.05%
49 fc_Linear 2048 1000 2049000 0.00MB 4,095,000 0.24%
===============================================================================================================================
total parameters quantity: 25,557,032
total memory: 109.69MB
total MAdd: 8,219,737,624
```
### validation mode
To make sure the result is correct, a validation mothod from https://github.com/sksq96/pytorch-summary is added.
To use this validation method, please install *torchsummary*
`pip3 install torchsummary`