tensorflow/models · error · ValueError

Wrong validation dataset size

Error message

Wrong validation dataset size {!r}

What it means

Companion check in add_trainer: num_val_examples is zero or negative. Validation steps per epoch are derived from this count; a non-positive value would skip or corrupt evaluation scheduling. Like the train-side check, the message interpolates experiment.task.train_data's repr instead of the number itself, which is misleading when reading the error.

Solutions

  1. Pass a positive num_val_examples (default YT8M_VAL_EXAMPLES)
  2. Count validation records for custom datasets and pass that number
  3. Check the yt8m_experiment caller for arguments passed in the wrong positional order
Defensive patterns

Strategy: validation

When it happens

Trigger: Thrown at official/projects/yt8m/configs/yt8m.py:231 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/a171985d86efbd22. Report an issue: GitHub.

Appendix: source

Thrown at official/projects/yt8m/configs/yt8m.py:231

  )
  gradient_clip_norm: float = 1.0


def add_trainer(
    experiment: cfg.ExperimentConfig,
    train_batch_size: int,
    eval_batch_size: int,
    learning_rate: float = 0.0001,
    train_epochs: int = 50,
    num_train_examples: int = YT8M_TRAIN_EXAMPLES,
    num_val_examples: int = YT8M_VAL_EXAMPLES,
) -> cfg.ExperimentConfig:
  """Adds and config a trainer to the experiment config."""
  if num_train_examples <= 0:
    raise ValueError('Wrong train dataset size {!r}'.format(
        experiment.task.train_data))
  if num_val_examples <= 0:
    raise ValueError('Wrong validation dataset size {!r}'.format(
        experiment.task.validation_data))
  experiment.task.train_data.global_batch_size = train_batch_size
  experiment.task.validation_data.global_batch_size = eval_batch_size
  steps_per_epoch = num_train_examples // train_batch_size
  steps_per_loop = 500
  experiment.trainer = cfg.TrainerConfig(
      steps_per_loop=steps_per_loop,
      summary_interval=steps_per_loop,
      checkpoint_interval=steps_per_loop,
      train_steps=train_epochs * steps_per_epoch,
      validation_steps=num_val_examples // eval_batch_size,
      validation_interval=steps_per_loop,
      optimizer_config=optimization.OptimizationConfig({
          'optimizer': {
              'type': 'adam',
              'adam': {}
          },
          'learning_rate': {

View on GitHub (pinned to e006f5f0d5)