tensorflow/models · error · ValueError

Pre-batch ({}) size is not equal to batch size ({})

Error message

Pre-batch ({}) size is not equal to batch size ({})

What it means

Error "Pre-batch ({}) size is not equal to batch size ({})" thrown in tensorflow/models.

Source

Thrown at official/recommendation/ncf_input_pipeline.py:35

import functools

# pylint: disable=g-bad-import-order
import tensorflow as tf, tf_keras
# pylint: enable=g-bad-import-order

from official.recommendation import constants as rconst
from official.recommendation import data_pipeline
from official.recommendation import movielens


def create_dataset_from_tf_record_files(input_file_pattern,
                                        pre_batch_size,
                                        batch_size,
                                        is_training=True,
                                        rebatch=False):
  """Creates dataset from (tf)records files for training/evaluation."""
  if pre_batch_size != batch_size:
    raise ValueError("Pre-batch ({}) size is not equal to batch "
                     "size ({})".format(pre_batch_size, batch_size))

  files = tf.data.Dataset.list_files(input_file_pattern, shuffle=is_training)

  dataset = files.interleave(
      tf.data.TFRecordDataset,
      cycle_length=16,
      num_parallel_calls=tf.data.experimental.AUTOTUNE)
  decode_fn = functools.partial(
      data_pipeline.DatasetManager.deserialize,
      batch_size=pre_batch_size,
      is_training=is_training)
  dataset = dataset.map(
      decode_fn, num_parallel_calls=tf.data.experimental.AUTOTUNE)

  if rebatch:
    # A workaround for TPU Pod evaluation dataset.
    # TODO (b/162341937) remove once it's fixed.

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/recommendation/ncf_input_pipeline.py:35 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/427c141c2c5e3e8b. Report an issue: GitHub.