Add topk_hyps, topk_ids, topk_lens, topk_scores tensors to `sample`. These features are required by `.BeamSearchDecodeOutput`. Args: sample: a NestedMap with `id`, `paddings`, and `logits` fields. is_tpu: whether inference is being run on TPU. Returns: sample with ad
(self, sample, is_tpu)
source not stored for this graph (policy: none)
no outgoing calls
no test coverage detected