I have an estimator that can train and eval using multiple GPU through specifying train_distribute and eval_distribute in config. But there's no predict_distribute argument in tf.estimator.RunConfig. How can I predict using multiple GPU?
I have an estimator that can train and eval using multiple GPU through specifying train_distribute and eval_distribute in config. But there's no predict_distribute argument in tf.estimator.RunConfig. How can I predict using multiple GPU?