ML Community Day is November 9! Join us for updates from TensorFlow, JAX, and more Learn more


Callback to save the Keras model or model weights at some frequency.

Inherits From: Callback

Used in the notebooks

Used in the guide Used in the tutorials

ModelCheckpoint callback is used in conjunction with training using to save a model or weights (in a checkpoint file) at some interval, so the model or weights can be loaded later to continue the training from the state saved.

A few options this callback provides include:

  • Whether to only keep the model that has achieved the "best performance" so far, or whether to save the model at the end of every epoch regardless of performance.
  • Definition of 'best'; which quantity to monitor and whether it should be maximized or minimized.
  • The frequency it should save at. Currently, the callback supports saving at the end of every epoch, or after a fixed number of training batches.
  • Whether only weights are saved, or the whole model is saved.


model.compile(loss=..., optimizer=...,

checkpoint_filepath = '/tmp/checkpoint'
model_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(

# Model weights are saved at the end of every epoch, if it's the best seen
# so far., callbacks=[model_checkpoint_callback])

# The model weights (that are considered the best) are loaded into the model.

filepath string or PathLike, path to save the model file. e.g. filepath = os.path.join(working_dir, 'ckpt', file_name). filepath can contain named formatting options, which will be filled the value of epoch and keys in logs (passed in on_epoch_end). For example: if filepath is weights.{epoch:02d}-{val_loss:.2f}.hdf5, then the model checkpoints will be saved with the epoch number and the validation loss in the filename. The directory of the filepath should not be reused by any other callbacks to avoid conflicts.
monitor The metric name to monitor. Typically the metrics are set by the Model.compile method. Note:

  • Prefix the name with "val_" to monitor validation metrics.
  • Use "loss" or "val_loss" to monitor the model's total loss.
  • If you specify metrics as strings, like "accuracy", pass the same string (with or without the "val_" prefix).
  • If you pass metrics.Metric objects, monitor should be set to
  • If you're not sure about the metric names you can check the contents of the history.history dictionary returned by history =
  • Multi-output models set additional prefixes on the metric names.
verbose verbosity mode, 0 or 1.