Select a specific dataset for training

[Elfsong]

seqio.TaskRegistry.add("task_1",
  source=seqio.FunctionDataSource(
	  dataset_fn=functools.partial(clef_dataset_fn, task="task_name", lang=lang),
	  splits=["train", "dev", ""test]
  ),
  preprocessors=[
	  c_preprocessor,
	  seqio.preprocessors.tokenize_and_append_eos,
  ],
  postprocess_fn=t5.data.postprocessors.lower_text,
  metric_fns=[c_metric],
  output_features=DEFAULT_OUTPUT_FEATURES,
)

May I ask which data split (train/dev/test) will be used for training by default? Can I name it specifically? Like using the dev dataset to train the model?

[Elfsong]

Can we use "utils.run.dataset_split" in the training phase?