tensorflow/models · error

Skip build head type: %s

Error message

Skip build head type: %s

What it means

Error "Skip build head type: %s" thrown in tensorflow/models.

Source

Thrown at official/projects/yt8m/modeling/yt8m_model.py:112

    self.backbone = backbone
    self.build_head()

  def build_head(self):
    logging.info("Build DbofModel with %s.", self._params.head.type)
    head_cfg = self._params.head.get()
    if self._params.head.type == "moe":
      normalizer_params = dict(
          synchronized=self._params.norm_activation.use_sync_bn,
          momentum=self._params.norm_activation.norm_momentum,
          epsilon=self._params.norm_activation.norm_epsilon,
      )
      aggregation_head = functools.partial(
          heads.MoeModel, normalizer_params=normalizer_params
      )
    elif self._params.head.type == "logistic":
      aggregation_head = heads.LogisticModel
    else:
      logging.warn("Skip build head type: %s", self._params.head.type)
      return

    l2_regularizer = (
        tf_keras.regularizers.l2(self._l2_weight_decay / 2.0)
        if self._l2_weight_decay
        else None
    )
    self.head = aggregation_head(
        vocab_size=self._num_classes,
        l2_regularizer=l2_regularizer,
        **head_cfg.as_dict(),
    )

  def get_config(self):
    return self._config_dict

  @classmethod
  def from_config(cls, config):  # pyrefly: ignore[bad-override]

View on GitHub (pinned to e006f5f0d5)

Solutions

1. Set params.head.type to a supported aggregation head: moe (Mixture of Experts) or logistic.
2. If a custom head type was intended, extend build_head() to construct it; otherwise the model is built without an aggregation head.

Example fix

# In the YT8M model config:
head:
  type: moe   # or logistic

When it happens

Trigger: Thrown at official/projects/yt8m/modeling/yt8m_model.py:112 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/552e083a904e4f04. Report an issue: GitHub.