I want to know if you uesd the weight decay in pytorch?

[solon37]

I am looking forward to hearing from you.

[fsschneider]

Whether the testproblems use weight decay (and if so for what part) is usually described in the comment of each testproblem. For example for the 3c3d network on cifar100:

DeepOBS/deepobs/pytorch/testproblems/cifar100_3c3d.py

Lines 21 to 22 in bfb1eb3

	- L2 regularization on the weights (but not the biases) with a default
	factor of 0.002

This is also listed in the documentation, e.g. here for the mlp on MNIST.

Table 1 of the DeepOBS paper has an overview table of the original testproblem and marked which ones use weight decay (WD).