I am running the trainer, with SGD (constant learning rate), and the learning rate reported is a square of what I am supplying. For example, when giving learning_rate = 2e-2, the reported learning_rate is ‘learning_rate’: 0.0004. Any ideas?
I am running the trainer, with SGD (constant learning rate), and the learning rate reported is a square of what I am supplying. For example, when giving learning_rate = 2e-2, the reported learning_rate is ‘learning_rate’: 0.0004. Any ideas?