Evaluation results on specific layers

I want to get the results of prediction, but not on the logits. I want to investigate the output of a specific layer.
Could you please tell me how?

You could register a forward hook to the specific layer and analyze the activation after the forward pass as shown in this example.

1 Like