tensorflow/models · error · ValueError

Fractional pooling is only supported for `pool_type`=`trunca

Error message

Fractional pooling is only supported for `pool_type`=`truncated_average`

What it means

Error "Fractional pooling is only supported for `pool_type`=`truncated_average`" thrown in tensorflow/models.

Source

Thrown at official/nlp/modeling/networks/funnel_transformer.py:393

    self._pooler_layer = tf_keras.layers.Dense(
        units=hidden_size,
        activation='tanh',
        kernel_initializer=tf_utils.clone_initializer(initializer),
        name='pooler_transform')
    if isinstance(pool_stride, int):
      # TODO(b/197133196): Pooling layer can be shared.
      pool_strides = [pool_stride] * num_layers
    else:
      if len(pool_stride) != num_layers:
        raise ValueError('Lengths of pool_stride and num_layers are not equal.')
      pool_strides = pool_stride

    is_fractional_pooling = False in [
        (1.0 * pool_stride).is_integer() for pool_stride in pool_strides
    ]
    if is_fractional_pooling and pool_type in [_MAX, _AVG]:
      raise ValueError(
          'Fractional pooling is only supported for'
          ' `pool_type`=`truncated_average`'
      )

    # TODO(crickwu): explore tf_keras.layers.serialize method.
    if pool_type == _MAX:
      pool_cls = tf_keras.layers.MaxPooling1D
    elif pool_type == _AVG:
      pool_cls = tf_keras.layers.AveragePooling1D
    elif pool_type == _TRUNCATED_AVG:
      # TODO(b/203665205): unpool_length should be implemented.
      if unpool_length != 0:
        raise ValueError('unpool_length is not supported by truncated_avg now.')
    else:
      raise ValueError('pool_type not supported.')

    if pool_type in (_MAX, _AVG):
      self._att_input_pool_layers = []

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/nlp/modeling/networks/funnel_transformer.py:393 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/30e69d3a4a176850. Report an issue: GitHub.