# Add L2 weight decay & adjust BN settings. model_config = model.get_config() for layer, layer_config in zip(model.layers, model_config['layers']): if hasattr(layer, 'kernel_regularizer'): regularizer = keras.regularizers.l2(args.wd) layer_config['config']['kernel_regularizer'] = \ {'class_name': regularizer.__class__.__name__, 'config': regularizer.get_config()} if type(layer) == keras.layers.BatchNormalization: layer_config['config']['momentum'] = 0.9 layer_config['config']['epsilon'] = 1e-5 model = keras.models.Model.from_config(model_config) # BytePS: adjust learning rate based on number of GPUs. opt = keras.optimizers.SGD(lr=args.base_lr * bps.size(), momentum=args.momentum) # BytePS: add BytePS Distributed Optimizer. opt = bps.DistributedOptimizer(opt, compression=compression) model.compile(loss=keras.losses.categorical_crossentropy, optimizer=opt, metrics=['accuracy', 'top_k_categorical_accuracy']) callbacks = [ # BytePS: broadcast initial variable states from rank 0 to all other processes. # This is necessary to ensure consistent initialization of all workers when # training is started with random weights or restored from a checkpoint. bps.callbacks.BroadcastGlobalVariablesCallback(0),
import byteps.keras as bps # BytePS: initialize BytePS. bps.init() # BytePS: pin GPU to be used to process local rank (one GPU per process) config = tf.ConfigProto() config.gpu_options.allow_growth = True config.gpu_options.visible_device_list = str(bps.local_rank()) K.set_session(tf.Session(config=config)) batch_size = 128 num_classes = 10 # BytePS: adjust number of epochs based on number of GPUs. epochs = int(math.ceil(12.0 / bps.size())) # Input image dimensions img_rows, img_cols = 28, 28 # The data, shuffled and split between train and test sets (x_train, y_train), (x_test, y_test) = mnist.load_data() if K.image_data_format() == 'channels_first': x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols) x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols) input_shape = (1, img_rows, img_cols) else: x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1) x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1) input_shape = (img_rows, img_cols, 1)
y_test = keras.utils.to_categorical(y_test, num_classes) model = Sequential() model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=input_shape)) model.add(Conv2D(64, (3, 3), activation='relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.25)) model.add(Flatten()) model.add(Dense(128, activation='relu')) model.add(Dropout(0.5)) model.add(Dense(num_classes, activation='softmax')) # BytePS: adjust learning rate based on number of GPUs. opt = keras.optimizers.Adadelta(lr=1.0 * bps.size()) # BytePS: add BytePS Distributed Optimizer. opt = bps.DistributedOptimizer(opt) model.compile(loss=keras.losses.categorical_crossentropy, optimizer=opt, metrics=['accuracy']) callbacks = [ # BytePS: broadcast initial variable states from rank 0 to all other processes. # This is necessary to ensure consistent initialization of all workers when # training is started with random weights or restored from a checkpoint. bps.callbacks.BroadcastGlobalVariablesCallback(0), # BytePS: average metrics among workers at the end of every epoch.