tf.keras.callbacks.BackupAndRestore

Callback to back up and restore the training state.

Inherits From: Callback

Used in the notebooks

Used in the guide Used in the tutorials

BackupAndRestore callback is intended to recover training from an interruption that has happened in the middle of a Model.fit execution, by backing up the training states in a temporary checkpoint file (with the help of a tf.train.CheckpointManager), at the end of each epoch. Each backup overwrites the previously written checkpoint file, so at any given time there is at most one such checkpoint file for backup/restoring purpose.

If training restarts before completion, the training state (which includes the Model weights and epoch number) is restored to the most recently saved state at the beginning of a new Model.fit run. At the completion of a Model.fit run, the temporary checkpoint file is deleted.

Note that the user is responsible to bring jobs back after the interruption. This callback is important for the backup and restore mechanism for fault tolerance purpose, and the model to be restored from a previous checkpoint is expected to be the same as the one used to back up. If user changes arguments passed to compile or fit, the checkpoint saved for fault tolerance can become invalid.

Note:

  1. This callback is not compatible with eager execution disabled.
  2. A checkpoint is saved at the end of each epoch. After restoring, Model.fit redoes any partial work during the unfinished epoch in which the training got restarted (so the work done before the interruption doesn't affect the final model state).
  3. This works for both single worker and multi-worker modes. When Model.fit is used with tf.distribute, it supports tf.distribute.MirroredStrategy, tf.distribute.MultiWorkerMirroredStrategy, tf.distribute.TPUStrategy, and tf.distribute.experimental.ParameterServerStrategy.

Example:

class InterruptingCallback(tf.keras.callbacks.Callback):
  def on_epoch_begin(self, epoch, logs=None):
    if epoch == 4:
      raise RuntimeError('Interrupting!')
callback = tf.keras.callbacks.BackupAndRestore(backup_dir="/tmp/backup")
model = tf.keras.models.Sequential([tf.keras.layers.Dense(10)])
model.compile(tf.keras.optimizers.SGD(), loss='mse')
try:
  model.fit(np.arange(100).reshape(5, 20), np.zeros(5), epochs=10,
            batch_size=1, callbacks=[callback, InterruptingCallback()],
            verbose=0)
except:
  pass
history = model.fit(np.arange(100).reshape(5, 20), np.zeros(5),
                    epochs=10, batch_size=1, callbacks=[callback],
                    verbose=0)
# Only 6 more epochs are run, since first training got interrupted at
# zero-indexed epoch 4, second training will continue from 4 to 9.
len(history.history['loss'])
6

Besides the option to save at the end of every epoch or every N steps, if you are doing distributed training with tf.distribute.MultiWorkerMirroredStrategy on Google Cloud Platform or Google Borg, you can also use the save_before_preemption argument to enable saving a checkpoint right before a worker gets preempted by other jobs and training gets interrupted. See tf.distribute.experimental.PreemptionCheckpointHandler for more details.

backup_dir String, path to store the checkpoint. e.g. backup_dir = os.path.join(working_dir, 'backup'). This is the directory in which the system stores temporary files to recover the model from jobs terminated unexpectedly. The directory cannot be reused elsewhere to store other files, e.g. by the BackupAndRestore callback of another training run, or by another callback (e.g. ModelCheckpoint) of the same training.
save_freq 'epoch', integer, or False. When set to 'epoch' the callback saves the checkpoint at the end of each epoch. When set to an integer, the callback saves the checkpoint every save_freq batches. Set save_freq to False if only using preemption checkpointing (with save_before_preemption=True).
delete_checkpoint Boolean, default to True. This BackupAndRestore callback works by saving a checkpoint to back up the training state. If delete_checkpoint=True, the checkpoint will be deleted after training is finished. Use False if you'd like to keep the checkpoint for future usage.
save_before_preemption A boolean value instructing whether to turn on the automatic checkpoint saving for preemption/maintenance events. This only supports tf.distribute.MultiWorkerMirroredStrategy on Google Cloud Platform or Google Borg for now.

Methods

set_model

View source

set_params

View source