Inference Sampler
sampler_inference(
sample_fn,
sample_shape,
sample_dtype = tf$int32,
end_fn,
next_inputs_fn = NULL,
...
)A callable that takes outputs and emits tensor sample_ids.
Either a list of integers, or a 1-D Tensor of type int32, the shape of the each sample in the batch returned by sample_fn.
the dtype of the sample returned by sample_fn.
A callable that takes sample_ids and emits a bool vector shaped [batch_size] indicating whether each sample is an end token.
(Optional) A callable that takes sample_ids and returns the next batch of inputs. If not provided, sample_ids is used as the next batch of inputs.
A list that contains other common arguments for layer creation.
None
A helper to use during inference with a custom sampling function.