tensorflow/models · error · ValueError

Inconsistent shapes in shard_tensors: first is {tensors[0].s

Error message

Inconsistent shapes in shard_tensors: first is {tensors[0].shape} and other is {tensor.shape}

What it means

Error "Inconsistent shapes in shard_tensors: first is {tensors[0].shape} and other is {tensor.shape}" thrown in tensorflow/models.

Source

Thrown at official/vision/modeling/layers/edgetpu.py:125

  Args:
    axis: axis to be used to split tensors
    block_size: block size to split tensors.
    tensors: list of tensors.

  Returns:
    List of shards, each shard has exactly one peace of each input tesnor.

  Raises:
    ValueError: if input tensors has different size of sharded dimension.
  """
  if not all(tensor.shape.is_fully_defined() for tensor in tensors):
    return [tensors]
  for validate_axis in range(axis + 1):
    consistent_length: int = tensors[0].shape[validate_axis]
    for tensor in tensors:
      if tensor.shape[validate_axis] != consistent_length:
        raise ValueError('Inconsistent shapes in shard_tensors: first is '
                         f'{tensors[0].shape} and other is {tensor.shape}')
  batch_size: int = tensors[0].shape[axis]
  if block_size >= batch_size:
    return [tensors]
  else:
    blocks = batch_size // block_size
    remainder = batch_size % block_size
    if remainder:
      tensor_parts = []
      for tensor in tensors:
        shape: tf.TensorShape = tensor.shape
        body: tf.Tensor = tf.slice(tensor, [0] * len(shape), [
            size if i != axis else blocks * block_size
            for i, size in enumerate(shape)
        ])
        tail: tf.Tensor = tf.slice(tensor, [
            0 if i != axis else (blocks * block_size)
            for i, _ in enumerate(shape)

View on GitHub (pinned to e006f5f0d5)

Solutions

1. Shard only tensors that all have identical shapes.
2. Pad or trim the tensors so every shard tensor has the same shape before calling shard_tensors.

When it happens

Trigger: Thrown at official/vision/modeling/layers/edgetpu.py:125 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/b467127590001219. Report an issue: GitHub.