{
  "dataset": "chiboard-1-sft",
  "serving_contract": "v2: an empty display column/slot means display == raw_pinyin. Consumers (training script, demo app, runtime) must send an empty <|reserved_7|> slot whenever the on-screen text equals the raw pinyin buffer; see README.md.",
  "boundary_semantics": "committed/active boundary = rolling decode-window retirement (plan.md section 2); no user-commit events. Cut preference: sentence-final punctuation > clause comma > word boundary.",
  "built_at": "2026-07-09T03:29:27",
  "random_seed": 20260703,
  "total_rows": 56810186,
  "split_counts": {
    "train": 54535518,
    "test": 1135059,
    "dev": 1139609
  },
  "dropped_exact_duplicates": 1191396,
  "length_distribution": {
    "2_8": {
      "target_share": 0.2447,
      "achieved_share": 0.256,
      "rows": 14541379,
      "ok": true
    },
    "9_30": {
      "target_share": 0.6232,
      "achieved_share": 0.6058,
      "rows": 34417375,
      "ok": true
    },
    "31_120": {
      "target_share": 0.1068,
      "achieved_share": 0.1115,
      "rows": 6334422,
      "ok": true
    },
    "121_500": {
      "target_share": 0.0212,
      "achieved_share": 0.0225,
      "rows": 1275407,
      "ok": true
    },
    "501_1500": {
      "target_share": 0.0035,
      "achieved_share": 0.0036,
      "rows": 206767,
      "ok": true
    },
    "1501_3500": {
      "target_share": 0.0006,
      "achieved_share": 0.0006,
      "rows": 34836,
      "ok": true
    }
  },
  "loss_tokens": {
    "tokenizer": "LiquidAI/LFM2.5-350M-Base",
    "total_sft_plus_reserve": 2055895946,
    "short_share": 0.4936,
    "per_bucket": {
      "121_500": 287113872,
      "1501_3500": 76109410,
      "2_8": 173677691,
      "31_120": 512930197,
      "501_1500": 164954634,
      "9_30": 841110142
    },
    "prompt_tokens_total": 3890996860,
    "loss_fraction": 0.3457,
    "empty_display_tokens_saved": 1737904853
  },
  "variant_distribution": {
    "joined_pinyin": {
      "target_share": 0.52,
      "achieved_share": 0.5487,
      "rows": 31172520,
      "ok": true
    },
    "mixed_chinese_pinyin": {
      "target_share": 0.15,
      "achieved_share": 0.1551,
      "rows": 8811285,
      "ok": true
    },
    "noisy_pinyin": {
      "target_share": 0.08,
      "achieved_share": 0.0771,
      "rows": 4380150,
      "ok": true
    },
    "mid_syllable": {
      "target_share": 0.08,
      "achieved_share": 0.0681,
      "rows": 3869283,
      "ok": true
    },
    "punctuation_removed": {
      "target_share": 0.07,
      "achieved_share": 0.0613,
      "rows": 3480239,
      "ok": true
    },
    "ascii_punctuation": {
      "target_share": 0.05,
      "achieved_share": 0.0431,
      "rows": 2449231,
      "ok": true
    },
    "syllable_spaced_pinyin": {
      "target_share": 0.03,
      "achieved_share": 0.028,
      "rows": 1589316,
      "ok": true
    },
    "already_chinese": {
      "target_share": 0.02,
      "achieved_share": 0.0186,
      "rows": 1058162,
      "ok": true
    }
  },
  "empty_display_share": 0.7968,
  "committed_context_share": 0.0738,
  "context_sentence_boundary_share": 1.0,
  "context_types": {
    "chunk": 3051518,
    "intra:sentence": 1139574
  },
  "retirement_boundary_intra_share": 0.2719,
  "retirement_cut_kinds": {
    "sentence": 1894699
  },
  "slice_rows": {
    "sft": 7017910,
    "reserve": 4583060
  },
  "active_length_histogram": {
    "1-9": 2685513,
    "10-19": 12245986,
    "20-39": 18781726,
    "40-69": 12922293,
    "70-100": 4482110,
    "101-160": 2481543,
    "161-300": 2316209,
    "301+": 894806
  },
  "active_window_share": 0.3064,
  "mixed_mode_counts": {
    "tail": 16019818,
    "edit": 3291800
  },
  "mixed_edit_share": 0.1705,
  "punct_style_counts": {
    "na": 22469862,
    "ascii": 21296348,
    "none": 3480239,
    "cjk": 3975404,
    "mixed": 4530171
  },
  "uv_counts": {
    "v_form": 5098493,
    "u_form": 566155
  },
  "noisy_rows": 5697476,
  "noise_rate_by_variant": {
    "joined_pinyin": 0.0,
    "mixed_chinese_pinyin": 0.0994,
    "noisy_pinyin": 1.0,
    "mid_syllable": 0.1141,
    "punctuation_removed": 0.0,
    "ascii_punctuation": 0.0,
    "syllable_spaced_pinyin": 0.0,
    "already_chinese": 0.0
  },
  "mixed_tail_length_hist_sampled": {
    "1-12": 103850,
    "121+": 3605,
    "13-40": 59936,
    "41-120": 8522
  },
  "hard_ambiguity_counts": {
    "changshi": 58936,
    "de_di_de": 22485533,
    "hangxing": 39425,
    "jiao": 1659343,
    "qishi": 430447,
    "shi": 15874452,
    "shijian": 797283,
    "ta": 3325425,
    "xian": 1513147,
    "yao": 5473893,
    "zai": 7761233,
    "zuo": 3081187
  },
  "coverage": {
    "top_n_words": 50000,
    "top_n_min_rows": 200,
    "top_n_meeting_quota": 50000,
    "top10k_words": 10000,
    "top10k_min_rows": 1000,
    "top10k_meeting_quota": 10000
  },
  "prompt_format": "<|startoftext|>{committed_context}<|reserved_6|>{raw_pinyin}<|reserved_7|>{display}<|reserved_8|>{target}<|im_end|> (plan.md section 2 v2)",
  "rows_globally_shuffled": true,
  "conversational_char_share": 0.4304,
  "per_source_chars": {
    "chinesewebtext2_hq": 467079375,
    "clapai_sentiment_zh": 18449321,
    "cmrc2018": 2679346,
    "dureader_retrieval_corpus": 260028027,
    "dureader_retrieval_queries": 783916,
    "feilongfl_news": 12216644,
    "lccc_dialogue": 9647571,
    "lccc_large": 407196903,
    "pd_dialogue": 6145469,
    "personal_dialog": 159860001,
    "somebreeze_news": 10005421
  },
  "per_source_rows": {
    "chinesewebtext2_hq": 7507058,
    "clapai_sentiment_zh": 578628,
    "cmrc2018": 51172,
    "dureader_retrieval_corpus": 4311996,
    "dureader_retrieval_queries": 85120,
    "feilongfl_news": 54260,
    "lccc_dialogue": 154574,
    "lccc_large": 30858374,
    "pd_dialogue": 88778,
    "personal_dialog": 12834156,
    "somebreeze_news": 286070
  },
  "source_manifest": [
    {
      "id": "lccc_large",
      "license": "mit",
      "url": "https://huggingface.co/datasets/thu-coai/lccc",
      "rows": 30858374
    },
    {
      "id": "personal_dialog",
      "license": "other (research-only; PersonalDialog, Zheng et al. 2019)",
      "url": "https://huggingface.co/datasets/silver/personal_dialog",
      "rows": 12834156
    },
    {
      "id": "dureader_retrieval_queries",
      "license": "apache-2.0",
      "url": "https://huggingface.co/datasets/zyznull/dureader-retrieval-corpus",
      "rows": 85120
    },
    {
      "id": "clapai_sentiment_zh",
      "license": "apache-2.0",
      "url": "https://huggingface.co/datasets/clapAI/MultiLingualSentiment",
      "rows": 578628
    },
    {
      "id": "somebreeze_news",
      "license": "apache-2.0",
      "url": "https://huggingface.co/datasets/somebreeze/Chinese-news-summery",
      "rows": 286070
    },
    {
      "id": "feilongfl_news",
      "license": "apache-2.0",
      "url": "https://huggingface.co/datasets/feilongfl/ChineseNewsSummary",
      "rows": 54260
    },
    {
      "id": "cmrc2018",
      "license": "cc-by-sa-4.0",
      "url": "https://huggingface.co/datasets/hfl/cmrc2018",
      "rows": 51172
    },
    {
      "id": "dureader_retrieval_corpus",
      "license": "apache-2.0",
      "url": "https://huggingface.co/datasets/zyznull/dureader-retrieval-corpus",
      "rows": 4311996
    },
    {
      "id": "chinesewebtext2_hq",
      "license": "apache-2.0",
      "url": "https://huggingface.co/datasets/Morton-Li/ChineseWebText2.0-HighQuality",
      "rows": 7507058
    },
    {
      "id": "lccc_dialogue",
      "license": "mit",
      "url": "https://huggingface.co/datasets/thu-coai/lccc",
      "derived_from": "lccc_large",
      "rows": 154574
    },
    {
      "id": "pd_dialogue",
      "license": "other (research-only; PersonalDialog, Zheng et al. 2019)",
      "url": "https://huggingface.co/datasets/silver/personal_dialog",
      "derived_from": "personal_dialog",
      "rows": 88778
    }
  ],
  "conversational_single_turn_share": 0.9945,
  "rejected_sources": [
    {
      "id": "lccc_base_train",
      "reason": "subset of lccc_large; would only add duplicate conversational turns"
    },
    {
      "id": "cc_cedict_rows",
      "reason": "dictionary headwords used for coverage accounting only; single-word rows are not realistic IME buffers"
    },
    {
      "id": "lucaswychan_dureader_hard_neg",
      "reason": "passages overlap dureader-retrieval-corpus; excluded to avoid near-duplicate leakage across doc ids"
    }
  ],
  "reserve": {
    "rows": 33578649,
    "directory": "/Users/bj/Desktop/Personal/Development/AI/Text Generation/Datasets/Chiboard-1/build_sft/reserve"
  },
  "plan_meta": {
    "total_planned_rows": 54358497,
    "planned_per_bucket": {
      "2_8": 15200867,
      "9_30": 32714072,
      "31_120": 4866417,
      "121_500": 1320723,
      "501_1500": 218745,
      "1501_3500": 37673
    },
    "reserve_planned_rows": 36200000,
    "reserve_planned_per_bucket": {
      "2_8": 10130580,
      "9_30": 21800480,
      "31_120": 3221520,
      "121_500": 877680,
      "501_1500": 144900,
      "1501_3500": 24840
    },
    "reserve_distinct_entries": {
      "2_8": 6668530,
      "9_30": 16877439,
      "31_120": 2819993,
      "121_500": 877680,
      "501_1500": 144900,
      "1501_3500": 24840
    },
    "slice_rows_sft": {
      "31_120": 4200126,
      "121_500": 3600639
    },
    "slice_rows_reserve": {
      "31_120": 2799874,
      "121_500": 2399361
    },
    "bucket_pool_sizes": {
      "2_8": 7300000,
      "9_30": 17401814,
      "31_120": 6009937,
      "121_500": 2701588,
      "501_1500": 420107,
      "1501_3500": 65023
    },
    "coverage_words": 50000,
    "coverage_deficient_words": 1487,
    "coverage_fill_rows": 37113,
    "ambiguity_fill_rows": 21384,
    "near_pair_groups": 8194,
    "near_pair_groups_per_term": {
      "de_di_de": 4486,
      "shi": 194,
      "ta": 2522,
      "zai": 477,
      "jiao": 149,
      "yao": 64,
      "zuo": 267,
      "xian": 30,
      "qishi": 3,
      "shijian": 2
    },
    "near_pair_included_rows": 0,
    "unreachable": {},
    "mult_cap": 4,
    "elapsed_seconds": 374.1
  },
  "qa_passed": true
}