I modified to unallocate GPU memory after the hold = False operation(train, evaluate, export, inference) except for the primary memory allocated by pytorch.
We will move the "hold" function to "waffle_store", so we will end it as a temporary measure for now.
Nevertheless, if you have a critical problem, please write a new issue.
resolve: #265
I modified to unallocate GPU memory after the
hold = False
operation(train
,evaluate
,export
,inference
) except for the primary memory allocated by pytorch.We will move the "hold" function to "waffle_store", so we will end it as a temporary measure for now. Nevertheless, if you have a critical problem, please write a new issue. resolve: #265