Closed Kirin-kun closed 5 years ago
This is an annoying error which has crept in when using masks. We're trying to find what causes it, but it's proving tricky.
It doesn't seem to impact training
I have traced this issue back to the Keras implementation of the Penalized Loss Function. I have now reverted that commit back to the pure Tensorflow implementation.
Unfortunately this issue seems to persist in TF even after reverting, so I will next look into how to get TF back to a state where this doesn't occur :/
Ok. If you still have issues after updating to the latest version, go into your Nvidia ComputeCache folder:
~/.nv/ComputeCache
(Linux)
C:\users\<USERNAME>\AppData\NVIDIA\ComputeCache
(windows)
and delete the contents.
Warning went away I think.
I'm trying the various plugins and with dfaker, on the training start, it gives a few warnings (errors?) , of the type:
The training still starts, so I don't know if there will be any consequences.
Can it be ignored or is it a concerning error?