tensorflow/models · error · ValueError

The input size (%d) is not a multiple of the number of atten

Error message

The input size (%d) is not a multiple of the number of attention heads (%d)

What it means

Error "The input size (%d) is not a multiple of the number of attention heads (%d)" thrown in tensorflow/models.

Source

Thrown at official/nlp/modeling/layers/tn_transformer_expand_condense.py:126

    if len(input_tensor_shape.as_list()) != 3:
      raise ValueError(
          "TNTransformerExpandCondense expects a three-dimensional input of "
          "shape [batch, sequence, width].")
    batch_size, sequence_length, hidden_size = input_tensor_shape

    if len(input_shape) == 2:
      mask_tensor_shape = tf.TensorShape(input_shape[1])
      expected_mask_tensor_shape = tf.TensorShape(
          [batch_size, sequence_length, sequence_length])
      if not expected_mask_tensor_shape.is_compatible_with(mask_tensor_shape):
        raise ValueError(
            "When passing a mask tensor to TNTransformerExpandCondense, the "
            "mask tensor must be of shape [batch, "
            "sequence_length, sequence_length] (here %s). Got a "
            "mask tensor of shape %s." %
            (expected_mask_tensor_shape, mask_tensor_shape))
    if hidden_size % self._num_heads != 0:
      raise ValueError(
          "The input size (%d) is not a multiple of the number of attention "
          "heads (%d)" % (hidden_size, self._num_heads))
    self._attention_head_size = int(hidden_size // self._num_heads)
    common_kwargs = dict(
        kernel_regularizer=self._kernel_regularizer,
        bias_regularizer=self._bias_regularizer,
        activity_regularizer=self._activity_regularizer,
        kernel_constraint=self._kernel_constraint,
        bias_constraint=self._bias_constraint)
    self._attention_layer = tf_keras.layers.MultiHeadAttention(
        num_heads=self._num_heads,
        key_dim=self._attention_head_size,
        dropout=self._attention_dropout_rate,
        use_bias=self._use_bias,
        kernel_initializer=self._attention_initializer,
        bias_initializer=tf_utils.clone_initializer(self._bias_initializer),
        name="self_attention",
        **common_kwargs)

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/nlp/modeling/layers/tn_transformer_expand_condense.py:126 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/622c15e3519ace2f. Report an issue: GitHub.