How to root cause - torch/autograd/__init__.py:xxx: UserWarning: Error detected in GeluBackward0. Traceback of forward call that caused the error

How do I dump more details or trace what exactly made autograd to fail ?

Could you please post the relevant code and the full error trace?

It is part of a lager model. The relevant past is as follows.

nn.Linear(...)
nn.GELU()
nn.Linear(...)
nn.Dropout(..)

Not able to reproduce in a smaller code snippet. How do I trace the autograd process ? is it possible to get a more detailed log of the autograd process ?

Will have to clean up the log a bit before I can post.