tensorflow/models · error · ValueError

`input_fn` should be a closure that returns a dataset.

Error message

`input_fn` should be a closure that returns a dataset.

What it means

Error "`input_fn` should be a closure that returns a dataset." thrown in tensorflow/models.

Source

Thrown at official/legacy/bert/model_training_utils.py:63

    logging.info('Saving model as TF checkpoint: %s', saved_path)
  else:
    # In multi worker training we need every worker to save checkpoint, because
    # variables can trigger synchronization on read and synchronization needs
    # all workers to participate. To avoid workers overriding each other we save
    # to a temporary directory on non-chief workers.
    tmp_dir = tempfile.mkdtemp()
    checkpoint.save(os.path.join(tmp_dir, 'ckpt'))
    tf.io.gfile.rmtree(tmp_dir)
  return


def _get_input_iterator(input_fn, strategy):
  """Returns distributed dataset iterator."""
  # When training with TPU pods, datasets needs to be cloned across
  # workers. Since Dataset instance cannot be cloned in eager mode, we instead
  # pass callable that returns a dataset.
  if not callable(input_fn):
    raise ValueError('`input_fn` should be a closure that returns a dataset.')
  iterator = iter(strategy.distribute_datasets_from_function(input_fn))
  return iterator


def _float_metric_value(metric):
  """Gets the value of a float-value keras metric."""
  return metric.result().numpy().astype(float)


def clip_by_global_norm_callback(grads_and_vars):
  """Performs gradient clipping."""
  grads, variables = zip(*grads_and_vars)
  (clipped_grads, _) = tf.clip_by_global_norm(grads, clip_norm=1.0)
  return zip(clipped_grads, variables)


def steps_to_run(current_step, steps_per_epoch, steps_per_loop):
  """Calculates steps to run on device."""

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/legacy/bert/model_training_utils.py:63 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/17c01f91bf5dcaac. Report an issue: GitHub.