-
Notifications
You must be signed in to change notification settings - Fork 124
Open
Description
Line 71 of utils.py
grads.append(0 if param.grad is None else param.grad + 0.)
should rewrite as:
grads.append(param-param if param.grad is None else param.grad + 0.)
The current implementation may cause bugs when there are unused layers in the model. To be specific, when a layer was set require_grad as true but doesn't participate in forward or backward participation, it's grad was set as float zero. It will trigger an error when torch.autograd checks the shape of grads. Detail can be seen in this discussion: #8
theseafaringturtle and ShunLu91
Metadata
Metadata
Assignees
Labels
No labels