tensorflow/models · error · TypeError

initial_cache element for key '%s' has dtype %s that does no

Error message

initial_cache element for key '%s' has dtype %s that does not match SequenceBeamSearch's dtype of %s. Value: %s

What it means

Error "initial_cache element for key '%s' has dtype %s that does not match SequenceBeamSearch's dtype of %s. Value: %s" thrown in tensorflow/models.

Source

Thrown at official/nlp/modeling/ops/beam_search.py:451

    finished_cond = tf.reduce_any(finished_flags, 1, name="finished_cond")
    seq_cond = _expand_to_same_rank(finished_cond, finished_seq)
    score_cond = _expand_to_same_rank(finished_cond, finished_scores)

    # Account for corner case where there are no finished sequences for a
    # particular batch item. In that case, return alive sequences for that batch
    # item.
    finished_seq = tf.where(seq_cond, finished_seq, alive_seq)
    finished_scores = tf.where(score_cond, finished_scores, alive_log_probs)
    return finished_seq, finished_scores

  def _create_initial_state(
      self, initial_ids, initial_cache, batch_size, constraint_mask=None
  ):
    """Return initial state dictionary and its shape invariants."""
    for key, value in initial_cache.items():
      for inner_value in tf.nest.flatten(value):
        if inner_value.dtype != self.dtype:
          raise TypeError(
              "initial_cache element for key '%s' has dtype %s that does not "
              "match SequenceBeamSearch's dtype of %s. Value: %s" %
              (key, inner_value.dtype.name, self.dtype.name, inner_value))

    # Current loop index (starts at 0)
    cur_index = tf.constant(0)

    # Create alive sequence with shape [batch_size, beam_size, 1]
    alive_seq = expand_to_beam_size(initial_ids, self.beam_size)
    alive_seq = tf.expand_dims(alive_seq, axis=2)
    if self.padded_decode:
      alive_seq = tf.tile(alive_seq, [1, 1, self.max_decode_length + 1])

    # Create tensor for storing initial log probabilities.
    # Assume initial_ids are prob 1.0
    initial_log_probs = tf.constant([[0.] + [-float("inf")] *
                                     (self.beam_size - 1)],
                                    dtype=self.dtype)

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/nlp/modeling/ops/beam_search.py:451 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/f4a3296321bd4f4c. Report an issue: GitHub.