PyTorch Freeze Some Layers or Parameters When Training – PyTorch Tutorial

When we are training a pytorch model, we may want to freeze some layers or parameter. In this tutorial, we will introduce you how to freeze and train.

Look at this model below:

import torch.nn as nn
from torch.autograd import Variable
import torch.optim as optim
class Net(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc1 = nn.Linear(2, 4)
        self.fc2 = nn.Linear(4, 3)
        self.out = nn.Linear(3, 1)
        self.out_act = nn.Sigmoid()

    def forward(self, inputs):
        a1 = self.fc1(inputs)
        a2 = self.fc2(a1)
        a3 = self.out(a2)
        y = self.out_act(a3)
        return y
model_1 = Net()

This code creates a model named model_1.

We can display all parameters in this model by model_1.state_dict()

params = model_1.state_dict()
print(params.keys())

We will see:

odict_keys(['fc1.weight', 'fc1.bias', 'fc2.weight', 'fc2.bias', 'out.weight', 'out.bias'])

You can know more on pytorch model.state_dict() in this tutorial:

Understand PyTorch model.state_dict() – PyTorch Tutorial

Then we can freeze some layers or parameters as follows:

for name, para in model_1.named_parameters():
    if name.startswith("fc1."):
        para.requires_grad = False

This code will freeze parameters that starts with “fc1.”

We can list all trainable parameters in pytorch model.

for name, para in model_1.named_parameters():
    print(name, para.requires_grad)

List All Trainable Variables in PyTorch – PyTorch Tutorial

We will get:

fc1.weight False
fc1.bias False
fc2.weight True
fc2.bias True
out.weight True
out.bias True

In order to train a model, we should create a optimizer for all trainable parameters.

Here is an example:

optimizer = optim.SGD(non_frozen_parameters, lr=0.1)

Then, we can start to train.