{
  "dataset": "UltraChat-Sinhala (UltraChat-200k machine-translated en->si, NLLB-200-3.3B)",
  "source": "HuggingFaceH4/ultrachat_200k",
  "split_method": "train/test membership taken verbatim from the original split by prompt_id",
  "formats": [
    "parquet (zstd, schema-compatible with ultrachat_200k)",
    "jsonl"
  ],
  "cleaning": [
    "NLLB ZWJ-conjunct repair",
    "MT masking-sentinel leak repair (in-place)",
    "dropped dialogues with empty turns",
    "deduplicated prompt_ids",
    "removed 1 prompt_id shared across SFT/GEN for global id-uniqueness"
  ],
  "generated": "2026-06-26",
  "splits": {
    "train_sft": {
      "records": 207831,
      "parquet_bytes": 708322973,
      "jsonl_bytes": 3275956541,
      "parquet_file": "train_sft.parquet",
      "jsonl_file": "train_sft.sinhala.jsonl"
    },
    "test_sft": {
      "records": 23106,
      "parquet_bytes": 78631342,
      "jsonl_bytes": 362942552,
      "parquet_file": "test_sft.parquet",
      "jsonl_file": "test_sft.sinhala.jsonl"
    },
    "train_gen": {
      "records": 255974,
      "parquet_bytes": 704284097,
      "jsonl_bytes": 3169354214,
      "parquet_file": "train_gen.parquet",
      "jsonl_file": "train_gen.sinhala.jsonl"
    },
    "test_gen": {
      "records": 28300,
      "parquet_bytes": 77517814,
      "jsonl_bytes": 348842209,
      "parquet_file": "test_gen.parquet",
      "jsonl_file": "test_gen.sinhala.jsonl"
    }
  }
}