tensorflow/models · error · AssertionError

Invalid dtype: %s

Error message

Invalid dtype: %s

What it means

Error "Invalid dtype: %s" thrown in tensorflow/models.

Source

Thrown at official/nlp/modeling/ops/beam_search.py:40

  """Returns a value close to infinity, but is still finite in `dtype`.

  This is useful to get a very large value that is still zero when multiplied by
  zero. The floating-point "Inf" value is NaN when multiplied by zero.

  Args:
    dtype: A dtype. The returned value will be finite when casted to this dtype.

  Returns:
    A very large value.
  """
  if dtype == "float32" or dtype == "bfloat16":
    return 1e7
  elif dtype == "float16":
    # Disable no-member lint error, as the linter thinks np.float16 does not
    # exist for some reason.
    return np.finfo(np.float16).max  # pylint: disable=no-member
  else:
    raise AssertionError("Invalid dtype: %s" % dtype)


class _StateKeys(object):
  """Keys to dictionary storing the state of the beam search loop."""

  # Variable storing the loop index.
  CUR_INDEX = "CUR_INDEX"

  # Top sequences that are alive for each batch item. Alive sequences are ones
  # that have not generated an EOS token. Sequences that reach EOS are marked as
  # finished and moved to the FINISHED_SEQ tensor.
  # Has shape [batch_size, beam_size, CUR_INDEX + 1]
  ALIVE_SEQ = "ALIVE_SEQ"
  # Log probabilities of each alive sequence. Shape [batch_size, beam_size]
  ALIVE_LOG_PROBS = "ALIVE_LOG_PROBS"
  # Dictionary of cached values for each alive sequence. The cache stores
  # the encoder output, attention bias, and the decoder attention output from
  # the previous iteration.

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/nlp/modeling/ops/beam_search.py:40 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/97b5b349fbaa550f. Report an issue: GitHub.