That is, we have a parameter tensor a with shape (10, 10).
We only want to train the weights for a[:3, :3].
What is the correct way of doing this? Seems SGD([a[:3, :3]], lr=0.01) does not work.
That is, we have a parameter tensor a with shape (10, 10).
We only want to train the weights for a[:3, :3].
What is the correct way of doing this? Seems SGD([a[:3, :3]], lr=0.01) does not work.