tensorflow/models · error · ValueError

best checkpoint metric comp must be one of higher, lower. Go

Error message

best checkpoint metric comp must be one of higher, lower. Got: {}

What it means

Error "best checkpoint metric comp must be one of higher, lower. Got: {}" thrown in tensorflow/models.

Source

Thrown at official/core/train_utils.py:154

  Orbit will support an API for checkpoint exporter. This class will be used
  together with orbit once this functionality is ready.
  """

  def __init__(self, export_dir: str, metric_name: str, metric_comp: str):
    """Initialization.

    Args:
      export_dir: The directory that will contain exported checkpoints.
      metric_name: Indicates which metric to look at, when determining which
        result is better. If eval_logs being passed to maybe_export_checkpoint
        is a nested dictionary, use `|` as a seperator for different layers.
      metric_comp: Indicates how to compare results. Either `lower` or `higher`.
    """
    self._export_dir = export_dir
    self._metric_name = metric_name.split('|')
    self._metric_comp = metric_comp
    if self._metric_comp not in ('lower', 'higher'):
      raise ValueError('best checkpoint metric comp must be one of '
                       'higher, lower. Got: {}'.format(self._metric_comp))
    tf.io.gfile.makedirs(os.path.dirname(self.best_ckpt_logs_path))
    self._best_ckpt_logs = self._maybe_load_best_eval_metric()
    self._checkpoint_manager = None

  def _get_checkpoint_manager(self, checkpoint):
    """Gets an existing checkpoint manager or creates a new one."""
    if self._checkpoint_manager is None or (self._checkpoint_manager.checkpoint
                                            != checkpoint):
      logging.info('Creates a new checkpoint manager.')
      self._checkpoint_manager = tf.train.CheckpointManager(
          checkpoint,
          directory=self._export_dir,
          max_to_keep=1,
          checkpoint_name=BEST_CHECKPOINT_NAME)

    return self._checkpoint_manager

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/core/train_utils.py:154 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of tensorflow/models@e006f5f0d5 (2026-08-24). Data as JSON: /api/errors/860e45ffdc36668c. Report an issue: GitHub.