train on GPU
littlealili opened this issue · comments
littlealili commented
Hello, if I want to run training on the GPU, how should I modify the program?
RuiLiFeng commented
If you only want to run on a single GPU, then no change is needed.
If you want to run on multi-gpus, then you may need to define your own training_loop, as tf.estimator is not supported well for multi-gpu training.