Does pytorch reset weights and biases when retraining a pretrain model

Hi there, I was wondering if for instance a pre-trained resnet50 model’s weights and biases would reset when training it with a new set of data.

You might want to look into-fine tuning a model.

When you fine-tune a model with another dataset, you take advantage that the pretrained weights mean something already and are not just random numbers.

When you do this, you want to take a small learning rate in order to achieve the desired result.

So, to answer your question, yeah the weights do change but they are not ‘reset’ to something random.

Hope this helps :slight_smile:

1 Like