Hello, I found that in the train_net.py, the initialization of the amp uses 'model, optimizer = amp.initialize(model, optimizer, opt_level=amp_opt_level)' and in the test_net.py, the initialization of the amp uses 'amp_handle = amp.init(enabled=use_mixed_precision, verbose=cfg.AMP_VERBOSE)' instead. What is the difference of them?
Hello, I found that in the train_net.py, the initialization of the amp uses 'model, optimizer = amp.initialize(model, optimizer, opt_level=amp_opt_level)' and in the test_net.py, the initialization of the amp uses 'amp_handle = amp.init(enabled=use_mixed_precision, verbose=cfg.AMP_VERBOSE)' instead. What is the difference of them?