import torch
import torch.nn as nn
m = nn.Linear(20, 30)
input = torch.randn(128, 3,20)
output = m(input)
print(output.size())
flops = (torch.prod(torch.LongTensor(list(output.size()))) \
* input[0].size(1)).item()
print((list(output.size())))
print(input[0].size(1))
print(flops)
官网对nn.Linear的介绍:
Applies a linear transformation to the incoming data:
y
=
x
A
T
+
b
y = xA^T + b
y=xAT+b
其函数签名为
torch.nn.Linear(in_features, out_features, bias=True)
其参数为:
in_features
- size of each input sample
out_features
-size of each output sample
bias
If set to false, the layer will not learn an additive bias.
其输入输出的维度为:- Input: ( N , ∗ , H i n ) (N, *, H_{in}) (N,∗,Hin), where *∗ means any number of additional dimensions and H i n H_{in} Hin=in_features
- Output: ( N , ∗ , H o u t ) (N, *, H_{out}) (N,∗,Hout), where all but the last dimension are the same shape as the input and H o u t H_{out} Hout=out_features
故代码运行得到的计算量FLOPS为 128×3×20×30 = 230400,即计算量FLOPs为batch_size×特征维度值×输入特征数×输出特征数。