Awesome
torchsummaryX
Improved visualization tool of torchsummary. Here, it visualizes kernel size, output shape, # params, and Mult-Adds. Also the torchsummaryX can handle RNN, Recursive NN, or model with multiple inputs.
Usage
pip install torchsummaryX and
from torchsummaryX import summary
summary(your_model, torch.zeros((1, 3, 224, 224)))
Args:
model(Module): Model to summarizex(Tensor): Input tensor of the model with [N, C, H, W] shape dtype and device have to match to the modelargs, kwargs: Other arguments used inmodel.forwardfunction
Examples
CNN for MNIST
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
self.conv2 = nn.Conv2d(10, 20, kernel_size=5)
self.conv2_drop = nn.Dropout2d()
self.fc1 = nn.Linear(320, 50)
self.fc2 = nn.Linear(50, 10)
def forward(self, x):
x = F.relu(F.max_pool2d(self.conv1(x), 2))
x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))
x = x.view(-1, 320)
x = F.relu(self.fc1(x))
x = F.dropout(x, training=self.training)
x = self.fc2(x)
return F.log_softmax(x, dim=1)
summary(Net(), torch.zeros((1, 1, 28, 28)))
=================================================================
Kernel Shape Output Shape Params Mult-Adds
Layer
0_conv1 [1, 10, 5, 5] [1, 10, 24, 24] 260.0 144.0k
1_conv2 [10, 20, 5, 5] [1, 20, 8, 8] 5.02k 320.0k
2_conv2_drop - [1, 20, 8, 8] - -
3_fc1 [320, 50] [1, 50] 16.05k 16.0k
4_fc2 [50, 10] [1, 10] 510.0 500.0
-----------------------------------------------------------------
Totals
Total params 21.84k
Trainable params 21.84k
Non-trainable params 0.0
Mult-Adds 480.5k
=================================================================
RNN
class Net(nn.Module):
def __init__(self,
vocab_size=20, embed_dim=300,
hidden_dim=512, num_layers=2):
super().__init__()
self.hidden_dim = hidden_dim
self.embedding = nn.Embedding(vocab_size, embed_dim)
self.encoder = nn.LSTM(embed_dim, hidden_dim,
num_layers=num_layers)
self.decoder = nn.Linear(hidden_dim, vocab_size)
def forward(self, x):
embed = self.embedding(x)
out, hidden = self.encoder(embed)
out = self.decoder(out)
out = out.view(-1, out.size(2))
return out, hidden
inputs = torch.zeros((100, 1), dtype=torch.long) # [length, batch_size]
summary(Net(), inputs)
===========================================================
Kernel Shape Output Shape Params Mult-Adds
Layer
0_embedding [300, 20] [100, 1, 300] 6000 6000
1_encoder - [100, 1, 512] 3768320 3760128
2_decoder [512, 20] [100, 1, 20] 10260 10240
-----------------------------------------------------------
Totals
Total params 3784580
Trainable params 3784580
Non-trainable params 0
Mult-Adds 3776368
===========================================================
Recursive NN
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(64, 64, 3, 1, 1)
def forward(self, x):
out = self.conv1(x)
out = self.conv1(out)
return out
summary(Net(), torch.zeros((1, 64, 28, 28)))
============================================================
Kernel Shape Output Shape Params Mult-Adds
Layer
0_conv1 [64, 64, 3, 3] [1, 64, 28, 28] 36.928k 28901376
1_conv1 [64, 64, 3, 3] [1, 64, 28, 28] - 28901376
------------------------------------------------------------
Totals
Total params 36.928k
Trainable params 36.928k
Non-trainable params 0.0
Mult-Adds 57.802752M
============================================================
Multiple arguments
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(64, 64, 3, 1, 1)
def forward(self, x, args1, args2):
out = self.conv1(x)
out = self.conv1(out)
return out
summary(Net(), torch.zeros((1, 64, 28, 28)), "args1", args2="args2")
Large models with long layer names
import torchvision
model = torchvision.models.resnet18()
summary(model, torch.zeros(4, 3, 224, 224))
=================================================================================================
Kernel Shape Output Shape \
Layer
0_conv1 [3, 64, 7, 7] [4, 64, 112, 112]
1_bn1 [64] [4, 64, 112, 112]
2_relu - [4, 64, 112, 112]
3_maxpool - [4, 64, 56, 56]
4_layer1.0.Conv2d_conv1 [64, 64, 3, 3] [4, 64, 56, 56]
5_layer1.0.BatchNorm2d_bn1 [64] [4, 64, 56, 56]
6_layer1.0.ReLU_relu - [4, 64, 56, 56]
7_layer1.0.Conv2d_conv2 [64, 64, 3, 3] [4, 64, 56, 56]
8_layer1.0.BatchNorm2d_bn2 [64] [4, 64, 56, 56]
9_layer1.0.ReLU_relu - [4, 64, 56, 56]
10_layer1.1.Conv2d_conv1 [64, 64, 3, 3] [4, 64, 56, 56]
11_layer1.1.BatchNorm2d_bn1 [64] [4, 64, 56, 56]
12_layer1.1.ReLU_relu - [4, 64, 56, 56]
13_layer1.1.Conv2d_conv2 [64, 64, 3, 3] [4, 64, 56, 56]
14_layer1.1.BatchNorm2d_bn2 [64] [4, 64, 56, 56]
15_layer1.1.ReLU_relu - [4, 64, 56, 56]
16_layer2.0.Conv2d_conv1 [64, 128, 3, 3] [4, 128, 28, 28]
17_layer2.0.BatchNorm2d_bn1 [128] [4, 128, 28, 28]
18_layer2.0.ReLU_relu - [4, 128, 28, 28]
19_layer2.0.Conv2d_conv2 [128, 128, 3, 3] [4, 128, 28, 28]
20_layer2.0.BatchNorm2d_bn2 [128] [4, 128, 28, 28]
21_layer2.0.downsample.Conv2d_0 [64, 128, 1, 1] [4, 128, 28, 28]
22_layer2.0.downsample.BatchNorm2d_1 [128] [4, 128, 28, 28]
23_layer2.0.ReLU_relu - [4, 128, 28, 28]
24_layer2.1.Conv2d_conv1 [128, 128, 3, 3] [4, 128, 28, 28]
25_layer2.1.BatchNorm2d_bn1 [128] [4, 128, 28, 28]
26_layer2.1.ReLU_relu - [4, 128, 28, 28]
27_layer2.1.Conv2d_conv2 [128, 128, 3, 3] [4, 128, 28, 28]
28_layer2.1.BatchNorm2d_bn2 [128] [4, 128, 28, 28]
29_layer2.1.ReLU_relu - [4, 128, 28, 28]
30_layer3.0.Conv2d_conv1 [128, 256, 3, 3] [4, 256, 14, 14]
31_layer3.0.BatchNorm2d_bn1 [256] [4, 256, 14, 14]
32_layer3.0.ReLU_relu - [4, 256, 14, 14]
33_layer3.0.Conv2d_conv2 [256, 256, 3, 3] [4, 256, 14, 14]
34_layer3.0.BatchNorm2d_bn2 [256] [4, 256, 14, 14]
35_layer3.0.downsample.Conv2d_0 [128, 256, 1, 1] [4, 256, 14, 14]
36_layer3.0.downsample.BatchNorm2d_1 [256] [4, 256, 14, 14]
37_layer3.0.ReLU_relu - [4, 256, 14, 14]
38_layer3.1.Conv2d_conv1 [256, 256, 3, 3] [4, 256, 14, 14]
39_layer3.1.BatchNorm2d_bn1 [256] [4, 256, 14, 14]
40_layer3.1.ReLU_relu - [4, 256, 14, 14]
41_layer3.1.Conv2d_conv2 [256, 256, 3, 3] [4, 256, 14, 14]
42_layer3.1.BatchNorm2d_bn2 [256] [4, 256, 14, 14]
43_layer3.1.ReLU_relu - [4, 256, 14, 14]
44_layer4.0.Conv2d_conv1 [256, 512, 3, 3] [4, 512, 7, 7]
45_layer4.0.BatchNorm2d_bn1 [512] [4, 512, 7, 7]
46_layer4.0.ReLU_relu - [4, 512, 7, 7]
47_layer4.0.Conv2d_conv2 [512, 512, 3, 3] [4, 512, 7, 7]
48_layer4.0.BatchNorm2d_bn2 [512] [4, 512, 7, 7]
49_layer4.0.downsample.Conv2d_0 [256, 512, 1, 1] [4, 512, 7, 7]
50_layer4.0.downsample.BatchNorm2d_1 [512] [4, 512, 7, 7]
51_layer4.0.ReLU_relu - [4, 512, 7, 7]
52_layer4.1.Conv2d_conv1 [512, 512, 3, 3] [4, 512, 7, 7]
53_layer4.1.BatchNorm2d_bn1 [512] [4, 512, 7, 7]
54_layer4.1.ReLU_relu - [4, 512, 7, 7]
55_layer4.1.Conv2d_conv2 [512, 512, 3, 3] [4, 512, 7, 7]
56_layer4.1.BatchNorm2d_bn2 [512] [4, 512, 7, 7]
57_layer4.1.ReLU_relu - [4, 512, 7, 7]
58_avgpool - [4, 512, 1, 1]
59_fc [512, 1000] [4, 1000]
Params Mult-Adds
Layer
0_conv1 9.408k 118.013952M
1_bn1 128.0 64.0
2_relu - -
3_maxpool - -
4_layer1.0.Conv2d_conv1 36.864k 115.605504M
5_layer1.0.BatchNorm2d_bn1 128.0 64.0
6_layer1.0.ReLU_relu - -
7_layer1.0.Conv2d_conv2 36.864k 115.605504M
8_layer1.0.BatchNorm2d_bn2 128.0 64.0
9_layer1.0.ReLU_relu - -
10_layer1.1.Conv2d_conv1 36.864k 115.605504M
11_layer1.1.BatchNorm2d_bn1 128.0 64.0
12_layer1.1.ReLU_relu - -
13_layer1.1.Conv2d_conv2 36.864k 115.605504M
14_layer1.1.BatchNorm2d_bn2 128.0 64.0
15_layer1.1.ReLU_relu - -
16_layer2.0.Conv2d_conv1 73.728k 57.802752M
17_layer2.0.BatchNorm2d_bn1 256.0 128.0
18_layer2.0.ReLU_relu - -
19_layer2.0.Conv2d_conv2 147.456k 115.605504M
20_layer2.0.BatchNorm2d_bn2 256.0 128.0
21_layer2.0.downsample.Conv2d_0 8.192k 6.422528M
22_layer2.0.downsample.BatchNorm2d_1 256.0 128.0
23_layer2.0.ReLU_relu - -
24_layer2.1.Conv2d_conv1 147.456k 115.605504M
25_layer2.1.BatchNorm2d_bn1 256.0 128.0
26_layer2.1.ReLU_relu - -
27_layer2.1.Conv2d_conv2 147.456k 115.605504M
28_layer2.1.BatchNorm2d_bn2 256.0 128.0
29_layer2.1.ReLU_relu - -
30_layer3.0.Conv2d_conv1 294.912k 57.802752M
31_layer3.0.BatchNorm2d_bn1 512.0 256.0
32_layer3.0.ReLU_relu - -
33_layer3.0.Conv2d_conv2 589.824k 115.605504M
34_layer3.0.BatchNorm2d_bn2 512.0 256.0
35_layer3.0.downsample.Conv2d_0 32.768k 6.422528M
36_layer3.0.downsample.BatchNorm2d_1 512.0 256.0
37_layer3.0.ReLU_relu - -
38_layer3.1.Conv2d_conv1 589.824k 115.605504M
39_layer3.1.BatchNorm2d_bn1 512.0 256.0
40_layer3.1.ReLU_relu - -
41_layer3.1.Conv2d_conv2 589.824k 115.605504M
42_layer3.1.BatchNorm2d_bn2 512.0 256.0
43_layer3.1.ReLU_relu - -
44_layer4.0.Conv2d_conv1 1.179648M 57.802752M
45_layer4.0.BatchNorm2d_bn1 1.024k 512.0
46_layer4.0.ReLU_relu - -
47_layer4.0.Conv2d_conv2 2.359296M 115.605504M
48_layer4.0.BatchNorm2d_bn2 1.024k 512.0
49_layer4.0.downsample.Conv2d_0 131.072k 6.422528M
50_layer4.0.downsample.BatchNorm2d_1 1.024k 512.0
51_layer4.0.ReLU_relu - -
52_layer4.1.Conv2d_conv1 2.359296M 115.605504M
53_layer4.1.BatchNorm2d_bn1 1.024k 512.0
54_layer4.1.ReLU_relu - -
55_layer4.1.Conv2d_conv2 2.359296M 115.605504M
56_layer4.1.BatchNorm2d_bn2 1.024k 512.0
57_layer4.1.ReLU_relu - -
58_avgpool - -
59_fc 513.0k 512.0k
-------------------------------------------------------------------------------------------------
Totals
Total params 11.689512M
Trainable params 11.689512M
Non-trainable params 0.0
Mult-Adds 1.814078144G
=================================================================================================