Just some quick comments:
-
You’re not even stating what you’re training to learn. For example, a basic classifier is generally easier to train than, say, a Variational Autoencoder.
-
Depending on your setup (i.e., what you’re trying to learn), are you sure you’re using the correct
loss_criterion. Most likely, but it is an easy trap to fall into. -
You didn’t post the code of you’re model. There are a bunch of things that can go wrong even if there’s not error and the model is at least somewhat training (e.g., see this post)
-
Your code looks “old”. From the docs: “The Variable API has been deprecated: Variables are no longer necessary to use autograd with tensors. Autograd automatically supports Tensors with
requires_gradset toTrue.”