one question

[day2nightya]

When I runing LOFGAN using checkpoints'resume in my own datasets, I meet Gradient explosion like the picture shows. Had you ever meet this question?
And another question, with LOFGAN running, the GPU'memory consumption goes bigger and bigger until memory overflow. How do you run it for 100000 epoches? I try to resume the model after memory overflow and go on training until 100000 epoches, but I only do once successfully which means the rest failed caused by Gradient explosion. Can you give me some advice?

[day2nightya]

Another thing deserved to mentioned is that the LOFGAN can run normally until 60000 epoches(memory overflow) without using checkpoints.