splits/split_stats.json

{
  "train_count": 10000000,
  "val_count": 500000,
  "test_count": 500000,
  "split_format": "hf_parquet_uint16_target_plus_padded_meta_v1",
  "subset_policy": {
    "train_count": 10000000,
    "val_count": 500000,
    "test_count": 500000
  },
  "rows_per_file": 32768,
  "compression": "zstd",
  "source_format": "compact_jsonl_gz_plus_target_uint16_x1000",
  "local_path_fields_removed": true
}