tensorflow/models · error · ValueError

Only 'backbone_projection' or 'backbone' can be used to init

Error message

Only 'backbone_projection' or 'backbone' can be used to initialize the model.

What it means

Error "Only 'backbone_projection' or 'backbone' can be used to initialize the model." thrown in tensorflow/models.

Source

Thrown at official/projects/simclr/modeling/multitask_model.py:113

    return tasks

  def initialize(self):
    """Loads the multi-task SimCLR model with a pretrained checkpoint."""
    ckpt_dir_or_file = self._config.init_checkpoint
    if tf.io.gfile.isdir(ckpt_dir_or_file):
      ckpt_dir_or_file = tf.train.latest_checkpoint(ckpt_dir_or_file)
    if not ckpt_dir_or_file:
      return

    logging.info('Loading pretrained %s', self._config.init_checkpoint_modules)
    if self._config.init_checkpoint_modules == 'backbone':
      pretrained_items = dict(backbone=self._backbone)
    elif self._config.init_checkpoint_modules == 'backbone_projection':
      pretrained_items = dict(
          backbone=self._backbone, projection_head=self._projection_head)
    else:
      raise ValueError(
          "Only 'backbone_projection' or 'backbone' can be used to "
          'initialize the model.')

    ckpt = tf.train.Checkpoint(**pretrained_items)
    status = ckpt.read(ckpt_dir_or_file)
    status.expect_partial().assert_existing_objects_matched()
    logging.info('Finished loading pretrained checkpoint from %s',
                 ckpt_dir_or_file)

  @property
  def checkpoint_items(self):
    """Returns a dictionary of items to be additionally checkpointed."""
    return dict(backbone=self._backbone, projection_head=self._projection_head)

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/projects/simclr/modeling/multitask_model.py:113 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/962095154e7d8e35. Report an issue: GitHub.