split_metadata.json

{
  "total_samples": 8013769,
  "validation_fraction": 0.001,
  "train_samples": 8005755,
  "validation_samples": 8014,
  "seed": 42
}