I like your flexible framework very much!
I’m just wondering is there any performance guide like https://www.tensorflow.org/performance/performance_guide
so that I can follow and accelerate my model, both training and testing.
One possible solution is to delete the tensors you don’t need any more, if the reference count goes to 0, then we can free some GPU memory resources? Will this also lead to some speed-up?
Thank you very much in advance!