tensorflow/models · error · ValueError

Invalid sequence length: {length} or shape: ({height, width}

Error message

Invalid sequence length: {length} or shape: ({height, width}).

What it means

Error "Invalid sequence length: {length} or shape: ({height, width})." thrown in tensorflow/models.

Source

Thrown at official/projects/maxvit/modeling/common_ops.py:249

  ret_shape = relative_position_tensor.shape.as_list()
  ret_shape[h_axis] = height * width
  ret_shape[h_axis + 1] = height * width
  reindexed_tensor = tf.reshape(reindexed_tensor, ret_shape)

  return reindexed_tensor


def float32_softmax(x: tf.Tensor, *args, **kwargs) -> tf.Tensor:
  y = tf.cast(tf.nn.softmax(tf.cast(x, tf.float32), *args, **kwargs), x.dtype)
  return y


def get_shape_from_length(length: int, height: int = 1, width: int = 1):
  """Gets input 2D shape from 1D sequence length."""
  input_height = int(math.sqrt(length * height // width))
  input_width = input_height * width // height
  if input_height * input_width != length:
    raise ValueError(
        f'Invalid sequence length: {length} or shape: ({height, width}).'
    )
  return (input_height, input_width)


def absolute_position_encoding(
    position: tf.Tensor, hidden_size: int, dtype=tf.float32) -> tf.Tensor:
  """Create absoulte position encoding."""
  position = tf.cast(position, dtype)
  half_hid = hidden_size // 2
  freq_seq = tf.cast(tf.range(half_hid), dtype=dtype)
  inv_freq = 1 / (10000 ** (freq_seq / half_hid))
  sinusoid = tf.einsum('S,D->SD', position, inv_freq)
  sin = tf.sin(sinusoid)
  cos = tf.cos(sinusoid)
  return tf.concat([sin, cos], axis=-1)

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/projects/maxvit/modeling/common_ops.py:249 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/ef05118efd7e164b. Report an issue: GitHub.