# Kuku Yalanji v20.0-full-candidate-corpus-gvn-3epoch-lr2e-5

Started UTC: 2026-07-02T13:39:22Z

Purpose: full 20,911-pair candidate-corpus continuation from v12 with tagged direct/reference Bible tasks, DB usage replay, and elder-shared sentence-pair replay.

Base model: /workspace/base_models/v12.0-tagged-direct-plus-reference-bible-gvn-token-4096row-25epoch-batch16/merged
Training file: /workspace/data/kuku_yalanji_v20_0_full_candidate_corpus_gvn/train.eng-gvn.jsonl
Validation file: /workspace/data/kuku_yalanji_v20_0_full_candidate_corpus_gvn/validation.eng-gvn.jsonl
Test file: /workspace/data/kuku_yalanji_v20_0_full_candidate_corpus_gvn/test.eng-gvn.jsonl
Decode: greedy, no anti-repeat constraints.
Batch: 8; grad accumulation: 2; epochs: 3; lr: 2e-5.
Training eval cap: validation=1024, test=1024.

## Dataset Manifest

```json
{
  "created_at": "2026-07-02",
  "purpose": "v20 full-corpus continuation set: all English->Kuku rows from the 20,911-pair Bible+sentence-candidate export, with tagged direct and reference-conditioned Bible tasks plus DB usage and elder-shared sentence pairs.",
  "full_corpus_dir": "/mnt/donto-data/donto-resources/research/translation-training/kuku-yalanji-runpod-2026-06-30/datasets/kuku_yalanji_ebible_parallel_with_sentence_candidates_v0.1.0",
  "v18_dir": "/mnt/donto-data/donto-resources/research/translation-training/kuku-yalanji-runpod-2026-06-30/prepared/v18.0-v10-plus-elder-sentence-pair",
  "output_dir": "/mnt/donto-data/donto-resources/research/translation-training/kuku-yalanji-runpod-2026-06-30/prepared/v20.0-full-candidate-corpus-gvn",
  "seed": "v20-full-candidate-corpus-2026-07-02",
  "usage_oversample": 4,
  "elder_sentence_pair_oversample": 16,
  "train_sample_rows_per_bible_task": 1024,
  "usage_split": {
    "train_rows": 365,
    "heldout_rows": 84,
    "train_word_ids": 298,
    "heldout_word_ids": 74,
    "word_id_intersection": 0
  },
  "bible_input_counts": {
    "train_source_rows": 16623,
    "validation_source_rows": 2153,
    "test_source_rows": 2135,
    "train_tiers": {
      "high_confidence_verse": 13270,
      "sentence_candidate": 3353
    },
    "validation_tiers": {
      "high_confidence_verse": 1670,
      "sentence_candidate": 483
    },
    "test_tiers": {
      "high_confidence_verse": 1657,
      "sentence_candidate": 478
    }
  },
  "input_counts": {
    "usage_train_rows": 365,
    "usage_heldout_rows": 84,
    "elder_sentence_pair_rows": 43
  },
  "output_counts": {
    "train_rows": 35394,
    "validation_rows": 4433,
    "test_rows": 4397,
    "bible_train_direct_rows": 16623,
    "bible_train_ref_rows": 16623,
    "bible_validation_direct_rows": 2153,
    "bible_validation_ref_rows": 2153,
    "bible_test_direct_rows": 2135,
    "bible_test_ref_rows": 2135,
    "usage_oversampled_rows": 1460,
    "elder_sentence_pair_oversampled_rows": 688,
    "eval_train_sample_rows": 2456
  },
  "outputs": [
    {
      "path": "bible/eval_train_sample.eng-gvn.jsonl",
      "rows": 2456,
      "sha256": "5c1779a753a019eeb0a578f7ccf7322b1d6f58b81d9642375563e13c170cd8dd"
    },
    {
      "path": "bible/test_direct.eng-gvn.jsonl",
      "rows": 2135,
      "sha256": "8ddc75608efed0f04af84238918f71cabde0b4b4ebc60cee6d9afef3765a05e5"
    },
    {
      "path": "bible/test_ref.eng-gvn.jsonl",
      "rows": 2135,
      "sha256": "475651407ca5a29d444ed40e43a6db5f31ec49e4297aac373a7bd952a1382212"
    },
    {
      "path": "bible/train_direct.eng-gvn.jsonl",
      "rows": 16623,
      "sha256": "83c7e7ce032032a8a9f8ca7fe626fb06e289d88d41bbcfb92f4a879df604013e"
    },
    {
      "path": "bible/train_ref.eng-gvn.jsonl",
      "rows": 16623,
      "sha256": "a3134c2c4c7b39e27953334c81a93d3bf7b136cc5589c1de0cf8c4330a1e0d78"
    },
    {
      "path": "bible/validation_direct.eng-gvn.jsonl",
      "rows": 2153,
      "sha256": "b27c2c70e8f7b0cc3f3f6b9f2ffb883d37aeb022f673162575930ea857a2cd1c"
    },
    {
      "path": "bible/validation_ref.eng-gvn.jsonl",
      "rows": 2153,
      "sha256": "23a87583062a96412715e4ad496efb9ea1f5c931234446455a3d1a9683409be0"
    },
    {
      "path": "db_usage/all_usage_examples.eng-gvn.jsonl",
      "rows": 449,
      "sha256": "bc5090a517b8e4d0d585134fdd0903b49575194d33016491dad45b04a3496c2d"
    },
    {
      "path": "db_usage/heldout_usage.eng-gvn.jsonl",
      "rows": 84,
      "sha256": "0155ba7e21de41da94dbe5b2270994a1152a31a14bcbde73140f4b0efebb6fc6"
    },
    {
      "path": "db_usage/train_usage.eng-gvn.jsonl",
      "rows": 365,
      "sha256": "dc3a9322381dafb1a786b22f707259d2b2d52cc1deb18fee55ed01ee4514b7bf"
    },
    {
      "path": "db_usage/train_usage_oversampled.eng-gvn.jsonl",
      "rows": 1460,
      "sha256": "723266dec43b7353c53b65093220e9993ab444da3a3b6dfca195decfee076a09"
    },
    {
      "path": "elder_sentence_pair/all_elder_sentence_pair.eng-gvn.jsonl",
      "rows": 43,
      "sha256": "29a9bfbfecbe535b56eda3bf0ecf255a17190b438186b60b999c78e1ee4854d2"
    },
    {
      "path": "elder_sentence_pair/train_elder_sentence_pair_oversampled.eng-gvn.jsonl",
      "rows": 688,
      "sha256": "d58f11d3dab2f15c4a1820ec46c7bd03a6855c6c436707b0493f36ec5e1f5b68"
    },
    {
      "path": "test.eng-gvn.jsonl",
      "rows": 4397,
      "sha256": "167791f206ef163f1ea6f0886e78302b7ef3e1e0f10872d25289c66d99c957ce"
    },
    {
      "path": "train.eng-gvn.jsonl",
      "rows": 35394,
      "sha256": "c5151fabb5cc660358e2e776da88c0d09726fab53a6c1081d276fe73a8ca312d"
    },
    {
      "path": "validation.eng-gvn.jsonl",
      "rows": 4433,
      "sha256": "6800de95417d7d2234d813ff267d41ab5d788ef1748943652f9d0d36d4cc20a3"
    }
  ],
  "training_note": "Train from the v12 merged model with target_lang=gvn_Latn, low LR, full train split, capped generated validation during training, and separate post-train evals for Bible direct/ref, DB usage, and elder-shared sentence pairs."
}
```

## Source-Length Preflight

```json
[
  {
    "file": "train.eng-gvn.jsonl",
    "rows": 35394,
    "min": 8,
    "max": 1288,
    "mean": 48.07826185229135,
    "median": 39.0,
    "p95": 78,
    "p99": 308,
    "max_source_length": 320,
    "over_max_source_length": 319,
    "first_over_max_source_length": [
      {
        "index": 184,
        "id": "6127839a-fee3-4bf5-9850-32279908e7c0:eng-gvn:bible_ref",
        "tokens": 358
      },
      {
        "index": 502,
        "id": "049a9d97-c488-4d6d-b2bb-b3c4584e9a9f:eng-gvn:translate",
        "tokens": 509
      },
      {
        "index": 527,
        "id": "b9af5116-6fd9-4f9d-ab2c-870ac276be13:eng-gvn:translate",
        "tokens": 649
      },
      {
        "index": 558,
        "id": "1d4108eb-411e-4500-8e0c-bd849e53beeb:eng-gvn:translate",
        "tokens": 453
      },
      {
        "index": 625,
        "id": "80ecc410-4980-4142-9c5c-31f973722891:eng-gvn:bible_ref",
        "tokens": 463
      },
      {
        "index": 656,
        "id": "b9fb7807-638e-4765-b03d-62f18d13bcea:eng-gvn:bible_ref",
        "tokens": 518
      },
      {
        "index": 813,
        "id": "94771133-f397-4a69-8ed6-efd444536068:eng-gvn:translate",
        "tokens": 423
      },
      {
        "index": 846,
        "id": "d0ea5b58-0eed-4f23-bcbf-389f0d45a8b8:eng-gvn:bible_ref",
        "tokens": 458
      },
      {
        "index": 1192,
        "id": "70d07ead-fc5c-4fa0-ae9d-3345d7002e32:eng-gvn:bible_ref",
        "tokens": 454
      },
      {
        "index": 1246,
        "id": "508e8f29-16dc-4717-b6db-29d858c9dcab:eng-gvn:bible_ref",
        "tokens": 479
      },
      {
        "index": 1345,
        "id": "f9e3821b-b167-4a22-87ca-72bd0980b4a5:eng-gvn:translate",
        "tokens": 323
      },
      {
        "index": 1387,
        "id": "3a661fc5-3b1b-4a47-af1a-6909cc64e4cf:eng-gvn:translate",
        "tokens": 356
      }
    ]
  },
  {
    "file": "validation.eng-gvn.jsonl",
    "rows": 4433,
    "min": 8,
    "max": 1637,
    "mean": 47.68441236183172,
    "median": 38,
    "p95": 76,
    "p99": 279,
    "max_source_length": 320,
    "over_max_source_length": 35,
    "first_over_max_source_length": [
      {
        "index": 23,
        "id": "f58ddd75-94e8-4e66-a22e-c2ef3eab9090:eng-gvn:translate",
        "tokens": 382
      },
      {
        "index": 77,
        "id": "40511cac-9c31-46a8-b44b-866e20cf92d2:eng-gvn:bible_ref",
        "tokens": 580
      },
      {
        "index": 221,
        "id": "d37e35ba-7fae-4486-8ee6-33c3d653216b:eng-gvn:translate",
        "tokens": 591
      },
      {
        "index": 629,
        "id": "40511cac-9c31-46a8-b44b-866e20cf92d2:eng-gvn:translate",
        "tokens": 569
      },
      {
        "index": 637,
        "id": "1f42d98e-342a-4bfe-aed3-cdb62ffd916b:eng-gvn:translate",
        "tokens": 484
      },
      {
        "index": 645,
        "id": "f80677d3-4a7c-46f7-9bd3-bec10ca53807:eng-gvn:bible_ref",
        "tokens": 401
      },
      {
        "index": 778,
        "id": "c81b0660-f3d3-4e15-b802-48111b41be77:eng-gvn:translate",
        "tokens": 506
      },
      {
        "index": 822,
        "id": "c81b0660-f3d3-4e15-b802-48111b41be77:eng-gvn:bible_ref",
        "tokens": 517
      },
      {
        "index": 892,
        "id": "f80677d3-4a7c-46f7-9bd3-bec10ca53807:eng-gvn:translate",
        "tokens": 391
      },
      {
        "index": 916,
        "id": "20e51bc8-638b-4d4f-a0d6-5adbb1f57a8f:eng-gvn:translate",
        "tokens": 357
      },
      {
        "index": 956,
        "id": "ca900d98-4ab8-40f8-bce7-bbbf953622ee:eng-gvn:bible_ref",
        "tokens": 530
      },
      {
        "index": 1104,
        "id": "a19df01a-3555-4a53-9922-992c9d052b15:eng-gvn:bible_ref",
        "tokens": 948
      }
    ]
  },
  {
    "file": "test.eng-gvn.jsonl",
    "rows": 4397,
    "min": 8,
    "max": 1310,
    "mean": 48.773254491698886,
    "median": 38,
    "p95": 72,
    "p99": 342,
    "max_source_length": 320,
    "over_max_source_length": 52,
    "first_over_max_source_length": [
      {
        "index": 217,
        "id": "a8267214-7e1f-459a-9afc-9167d722493a:eng-gvn:bible_ref",
        "tokens": 1242
      },
      {
        "index": 220,
        "id": "d7a96d3b-0c79-4fb4-ab1d-17dfc5805651:eng-gvn:translate",
        "tokens": 376
      },
      {
        "index": 455,
        "id": "968c8899-2c62-41ea-aea2-c6f775e9d55d:eng-gvn:bible_ref",
        "tokens": 332
      },
      {
        "index": 647,
        "id": "143a99eb-caca-4c98-9ad1-b75fcec94e33:eng-gvn:translate",
        "tokens": 963
      },
      {
        "index": 701,
        "id": "b4dd7d3e-ed87-485b-8411-1fbd53ee7327:eng-gvn:translate",
        "tokens": 777
      },
      {
        "index": 721,
        "id": "968c8899-2c62-41ea-aea2-c6f775e9d55d:eng-gvn:translate",
        "tokens": 321
      },
      {
        "index": 731,
        "id": "bde7cded-f583-4d1d-830c-e8b5870d0750:eng-gvn:bible_ref",
        "tokens": 347
      },
      {
        "index": 801,
        "id": "29873054-3db7-4f3a-a86d-ff0cb9ec1c51:eng-gvn:translate",
        "tokens": 342
      },
      {
        "index": 820,
        "id": "ce1e4246-d16b-47fb-ae37-e90f9994b3d3:eng-gvn:translate",
        "tokens": 428
      },
      {
        "index": 844,
        "id": "5bee9ff8-835f-4cee-bf57-c0b37a9126b9:eng-gvn:bible_ref",
        "tokens": 342
      },
      {
        "index": 850,
        "id": "2ef343f3-dd2f-4ddf-8eb9-a714444f084e:eng-gvn:bible_ref",
        "tokens": 658
      },
      {
        "index": 1181,
        "id": "0c2e087a-a0d0-474a-aaa4-6e0697f19c81:eng-gvn:translate",
        "tokens": 467
      }
    ]
  },
  {
    "file": "bible/eval_train_sample.eng-gvn.jsonl",
    "rows": 2456,
    "min": 10,
    "max": 1065,
    "mean": 47.03745928338762,
    "median": 39.0,
    "p95": 77,
    "p99": 247,
    "max_source_length": 320,
    "over_max_source_length": 16,
    "first_over_max_source_length": [
      {
        "index": 1,
        "id": "eed1fb34-c2c0-4f66-bb87-038f08f83970:eng-gvn:bible_ref",
        "tokens": 753
      },
      {
        "index": 35,
        "id": "1abdf4f2-9e13-4488-a765-c8d8e36cbf9a:eng-gvn:bible_ref",
        "tokens": 390
      },
      {
        "index": 93,
        "id": "e998361c-fb91-4dc3-87ac-843d56008723:eng-gvn:translate",
        "tokens": 492
      },
      {
        "index": 358,
        "id": "585d283e-b5f3-4302-a85a-3d217fa56844:eng-gvn:translate",
        "tokens": 1065
      },
      {
        "index": 361,
        "id": "0c4af18b-888e-47e3-bfc3-d49807f519b4:eng-gvn:translate",
        "tokens": 512
      },
      {
        "index": 374,
        "id": "731745b4-7566-4f93-a221-0237e8a862f8:eng-gvn:bible_ref",
        "tokens": 530
      },
      {
        "index": 572,
        "id": "a55ebca9-f109-4b17-9b8a-2e8a9487fb59:eng-gvn:translate",
        "tokens": 438
      },
      {
        "index": 660,
        "id": "7d9556da-5a47-4f36-a6b6-a3479a698c95:eng-gvn:bible_ref",
        "tokens": 322
      },
      {
        "index": 812,
        "id": "80404513-ba94-4089-8bb8-1180a18f6531:eng-gvn:bible_ref",
        "tokens": 606
      },
      {
        "index": 1276,
        "id": "ab6ca40c-d7f4-42d2-9255-53d018762711:eng-gvn:bible_ref",
        "tokens": 387
      },
      {
        "index": 1277,
        "id": "3aa08da0-faaf-4d03-919c-0ffbd23de52e:eng-gvn:bible_ref",
        "tokens": 373
      },
      {
        "index": 1501,
        "id": "5af52740-37e1-416b-8496-be07dd263df8:eng-gvn:bible_ref",
        "tokens": 489
      }
    ]
  },
  {
    "file": "bible/validation_direct.eng-gvn.jsonl",
    "rows": 2153,
    "min": 8,
    "max": 1627,
    "mean": 42.76962378077102,
    "median": 32,
    "p95": 71,
    "p99": 274,
    "max_source_length": 320,
    "over_max_source_length": 17,
    "first_over_max_source_length": [
      {
        "index": 110,
        "id": "40511cac-9c31-46a8-b44b-866e20cf92d2:eng-gvn:translate",
        "tokens": 569
      },
      {
        "index": 111,
        "id": "d37e35ba-7fae-4486-8ee6-33c3d653216b:eng-gvn:translate",
        "tokens": 591
      },
      {
        "index": 117,
        "id": "47a7de8c-26db-41a8-811a-a0a39029909f:eng-gvn:translate",
        "tokens": 1626
      },
      {
        "index": 118,
        "id": "972f954a-b438-4f32-80fe-fc38f10edeff:eng-gvn:translate",
        "tokens": 1627
      },
      {
        "index": 150,
        "id": "456cb830-5747-43c7-9a4c-eaee365c3e2e:eng-gvn:translate",
        "tokens": 917
      },
      {
        "index": 151,
        "id": "a19df01a-3555-4a53-9922-992c9d052b15:eng-gvn:translate",
        "tokens": 937
      },
      {
        "index": 269,
        "id": "1f42d98e-342a-4bfe-aed3-cdb62ffd916b:eng-gvn:translate",
        "tokens": 484
      },
      {
        "index": 270,
        "id": "976a429f-e03d-4c40-956c-86efec588aa5:eng-gvn:translate",
        "tokens": 501
      },
      {
        "index": 700,
        "id": "c81b0660-f3d3-4e15-b802-48111b41be77:eng-gvn:translate",
        "tokens": 506
      },
      {
        "index": 701,
        "id": "ca900d98-4ab8-40f8-bce7-bbbf953622ee:eng-gvn:translate",
        "tokens": 519
      },
      {
        "index": 781,
        "id": "4d499898-f8e5-4785-8295-59516f30ca31:eng-gvn:translate",
        "tokens": 324
      },
      {
        "index": 785,
        "id": "f80677d3-4a7c-46f7-9bd3-bec10ca53807:eng-gvn:translate",
        "tokens": 391
      }
    ]
  },
  {
    "file": "bible/validation_ref.eng-gvn.jsonl",
    "rows": 2153,
    "min": 20,
    "max": 1637,
    "mean": 53.22991175104505,
    "median": 42,
    "p95": 82,
    "p99": 285,
    "max_source_length": 320,
    "over_max_source_length": 18,
    "first_over_max_source_length": [
      {
        "index": 110,
        "id": "40511cac-9c31-46a8-b44b-866e20cf92d2:eng-gvn:bible_ref",
        "tokens": 580
      },
      {
        "index": 111,
        "id": "d37e35ba-7fae-4486-8ee6-33c3d653216b:eng-gvn:bible_ref",
        "tokens": 602
      },
      {
        "index": 117,
        "id": "47a7de8c-26db-41a8-811a-a0a39029909f:eng-gvn:bible_ref",
        "tokens": 1636
      },
      {
        "index": 118,
        "id": "972f954a-b438-4f32-80fe-fc38f10edeff:eng-gvn:bible_ref",
        "tokens": 1637
      },
      {
        "index": 150,
        "id": "456cb830-5747-43c7-9a4c-eaee365c3e2e:eng-gvn:bible_ref",
        "tokens": 928
      },
      {
        "index": 151,
        "id": "a19df01a-3555-4a53-9922-992c9d052b15:eng-gvn:bible_ref",
        "tokens": 948
      },
      {
        "index": 269,
        "id": "1f42d98e-342a-4bfe-aed3-cdb62ffd916b:eng-gvn:bible_ref",
        "tokens": 494
      },
      {
        "index": 270,
        "id": "976a429f-e03d-4c40-956c-86efec588aa5:eng-gvn:bible_ref",
        "tokens": 511
      },
      {
        "index": 700,
        "id": "c81b0660-f3d3-4e15-b802-48111b41be77:eng-gvn:bible_ref",
        "tokens": 517
      },
      {
        "index": 701,
        "id": "ca900d98-4ab8-40f8-bce7-bbbf953622ee:eng-gvn:bible_ref",
        "tokens": 530
      },
      {
        "index": 781,
        "id": "4d499898-f8e5-4785-8295-59516f30ca31:eng-gvn:bible_ref",
        "tokens": 334
      },
      {
        "index": 782,
        "id": "9c4fc003-a550-4af3-8256-477813a33c9e:eng-gvn:bible_ref",
        "tokens": 325
      }
    ]
  },
  {
    "file": "bible/test_direct.eng-gvn.jsonl",
    "rows": 2135,
    "min": 8,
    "max": 1299,
    "mean": 43.90257611241218,
    "median": 32,
    "p95": 66,
    "p99": 334,
    "max_source_length": 320,
    "over_max_source_length": 26,
    "first_over_max_source_length": [
      {
        "index": 130,
        "id": "a8267214-7e1f-459a-9afc-9167d722493a:eng-gvn:translate",
        "tokens": 1231
      },
      {
        "index": 131,
        "id": "d1d8eca5-828b-41d0-a519-33a22a15c5c9:eng-gvn:translate",
        "tokens": 1299
      },
      {
        "index": 167,
        "id": "29873054-3db7-4f3a-a86d-ff0cb9ec1c51:eng-gvn:translate",
        "tokens": 342
      },
      {
        "index": 168,
        "id": "5bee9ff8-835f-4cee-bf57-c0b37a9126b9:eng-gvn:translate",
        "tokens": 331
      },
      {
        "index": 173,
        "id": "968c8899-2c62-41ea-aea2-c6f775e9d55d:eng-gvn:translate",
        "tokens": 321
      },
      {
        "index": 174,
        "id": "c72fa4e3-6316-4a8e-a1c4-dc1da15eae8f:eng-gvn:translate",
        "tokens": 334
      },
      {
        "index": 175,
        "id": "42ba0506-f8e2-4e98-8f87-16d5e9aa707c:eng-gvn:translate",
        "tokens": 650
      },
      {
        "index": 176,
        "id": "c2090500-a9d2-4960-9bd2-7a98934614ab:eng-gvn:translate",
        "tokens": 645
      },
      {
        "index": 300,
        "id": "edef32de-6e9e-45ff-acc8-bd47e8830a0a:eng-gvn:translate",
        "tokens": 759
      },
      {
        "index": 301,
        "id": "b4dd7d3e-ed87-485b-8411-1fbd53ee7327:eng-gvn:translate",
        "tokens": 777
      },
      {
        "index": 308,
        "id": "3f56b3d3-541b-4992-8893-8ac4856090ff:eng-gvn:translate",
        "tokens": 528
      },
      {
        "index": 309,
        "id": "8fed6c32-6331-4672-bcad-c7ff317bd247:eng-gvn:translate",
        "tokens": 514
      }
    ]
  },
  {
    "file": "bible/test_ref.eng-gvn.jsonl",
    "rows": 2135,
    "min": 20,
    "max": 1310,
    "mean": 54.34473067915691,
    "median": 42,
    "p95": 77,
    "p99": 345,
    "max_source_length": 320,
    "over_max_source_length": 26,
    "first_over_max_source_length": [
      {
        "index": 130,
        "id": "a8267214-7e1f-459a-9afc-9167d722493a:eng-gvn:bible_ref",
        "tokens": 1242
      },
      {
        "index": 131,
        "id": "d1d8eca5-828b-41d0-a519-33a22a15c5c9:eng-gvn:bible_ref",
        "tokens": 1310
      },
      {
        "index": 167,
        "id": "29873054-3db7-4f3a-a86d-ff0cb9ec1c51:eng-gvn:bible_ref",
        "tokens": 353
      },
      {
        "index": 168,
        "id": "5bee9ff8-835f-4cee-bf57-c0b37a9126b9:eng-gvn:bible_ref",
        "tokens": 342
      },
      {
        "index": 173,
        "id": "968c8899-2c62-41ea-aea2-c6f775e9d55d:eng-gvn:bible_ref",
        "tokens": 332
      },
      {
        "index": 174,
        "id": "c72fa4e3-6316-4a8e-a1c4-dc1da15eae8f:eng-gvn:bible_ref",
        "tokens": 345
      },
      {
        "index": 175,
        "id": "42ba0506-f8e2-4e98-8f87-16d5e9aa707c:eng-gvn:bible_ref",
        "tokens": 661
      },
      {
        "index": 176,
        "id": "c2090500-a9d2-4960-9bd2-7a98934614ab:eng-gvn:bible_ref",
        "tokens": 656
      },
      {
        "index": 300,
        "id": "edef32de-6e9e-45ff-acc8-bd47e8830a0a:eng-gvn:bible_ref",
        "tokens": 770
      },
      {
        "index": 301,
        "id": "b4dd7d3e-ed87-485b-8411-1fbd53ee7327:eng-gvn:bible_ref",
        "tokens": 788
      },
      {
        "index": 308,
        "id": "3f56b3d3-541b-4992-8893-8ac4856090ff:eng-gvn:bible_ref",
        "tokens": 539
      },
      {
        "index": 309,
        "id": "8fed6c32-6331-4672-bcad-c7ff317bd247:eng-gvn:bible_ref",
        "tokens": 525
      }
    ]
  },
  {
    "file": "db_usage/heldout_usage.eng-gvn.jsonl",
    "rows": 84,
    "min": 33,
    "max": 53,
    "mean": 40.166666666666664,
    "median": 39.0,
    "p95": 48,
    "p99": 51,
    "max_source_length": 320,
    "over_max_source_length": 0,
    "first_over_max_source_length": []
  },
  {
    "file": "elder_sentence_pair/all_elder_sentence_pair.eng-gvn.jsonl",
    "rows": 43,
    "min": 13,
    "max": 62,
    "mean": 30.790697674418606,
    "median": 30,
    "p95": 44,
    "p99": 51,
    "max_source_length": 320,
    "over_max_source_length": 0,
    "first_over_max_source_length": []
  }
]
```

## GPU

```text
Thu Jul  2 13:39:36 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.124.06             Driver Version: 570.124.06     CUDA Version: 12.8     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA L40S                    On  |   00000000:84:00.0 Off |                    0 |
| N/A   35C    P8             36W /  350W |       1MiB /  46068MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+
```

Finished UTC: 2026-07-02T16:44:15Z

## Training Manifest

```json
{
  "model_id": "mobtranslate/kuku-yalanji-nllb-lora",
  "version": "0.1.0",
  "created_at": "2026-07-02T15:51:37.641886+00:00",
  "base_model": "/workspace/base_models/v12.0-tagged-direct-plus-reference-bible-gvn-token-4096row-25epoch-batch16/merged",
  "direction": "eng-gvn",
  "source_lang": "eng_Latn",
  "target_lang": "gvn_Latn",
  "target_lang_init_from": null,
  "source_lang_token_id": 256047,
  "target_lang_token_id": 256204,
  "dataset": {
    "train_file": "/workspace/data/kuku_yalanji_v20_0_full_candidate_corpus_gvn/train.eng-gvn.jsonl",
    "validation_file": "/workspace/data/kuku_yalanji_v20_0_full_candidate_corpus_gvn/validation.eng-gvn.jsonl",
    "test_file": "/workspace/data/kuku_yalanji_v20_0_full_candidate_corpus_gvn/test.eng-gvn.jsonl",
    "split_rows": {
      "train": 35394,
      "validation": 1024,
      "test": 1024
    },
    "max_train_samples": null,
    "max_validation_samples": 1024,
    "max_test_samples": 1024,
    "shuffle_before_cap": false
  },
  "training_args": {
    "epochs": 3.0,
    "batch_size": 8,
    "gradient_accumulation_steps": 2,
    "learning_rate": 2e-05,
    "lora_r": 64,
    "lora_alpha": 128,
    "lora_dropout": 0.0,
    "lora_target_modules": [
      "q_proj",
      "k_proj",
      "v_proj",
      "out_proj",
      "fc1",
      "fc2"
    ],
    "modules_to_save": [],
    "seed": 42,
    "ensure_weight_tying": true,
    "effective_ensure_weight_tying": true,
    "generation_num_beams": 1,
    "generation_no_repeat_ngram_size": 0,
    "generation_repetition_penalty": 1.0,
    "generation_length_penalty": 1.0,
    "save_total_limit": 2,
    "gradient_checkpointing": false
  },
  "metrics": {
    "train": {
      "train_runtime": 6767.6911,
      "train_samples_per_second": 15.69,
      "train_steps_per_second": 0.981,
      "total_flos": 8.119575264288768e+16,
      "train_loss": 1.4851415328335662,
      "epoch": 2.998870056497175
    },
    "validation": {
      "eval_loss": 1.5804394483566284,
      "eval_bleu": 13.406777236161425,
      "eval_chrf": 39.882101420536515,
      "eval_runtime": 556.9744,
      "eval_samples_per_second": 1.839,
      "eval_steps_per_second": 0.23,
      "epoch": 2.998870056497175
    },
    "test": {
      "test_loss": 1.5612670183181763,
      "test_bleu": 13.157394044826804,
      "test_chrf": 40.2505720878999,
      "test_runtime": 539.4346,
      "test_samples_per_second": 1.898,
      "test_steps_per_second": 0.237,
      "epoch": 2.998870056497175
    }
  },
  "artifacts": {
    "adapter_dir": "/workspace/models/kuku-yalanji-nllb-lora/v20.0-full-candidate-corpus-gvn-3epoch-lr2e-5/adapter",
    "merged_dir": "/workspace/models/kuku-yalanji-nllb-lora/v20.0-full-candidate-corpus-gvn-3epoch-lr2e-5/merged"
  }
}
```

## Post-Eval Analysis

```json
{
  "train_sample": {
    "metrics": {
      "bleu": 23.34846437322879,
      "chrf": 47.09634456556977,
      "rows": 2456,
      "num_beams": 1,
      "no_repeat_ngram_size": 0,
      "repetition_penalty": 1.0,
      "length_penalty": 1.0
    },
    "rows": 2456,
    "exact": 170,
    "exact_rate": 0.06921824104234528,
    "empty": 0,
    "length_ratio": {
      "mean": 1.0066174314310394,
      "median": 1.0,
      "min": 0.08235294117647059,
      "max": 17.4
    },
    "max_token_repeat_share": {
      "mean": 0.16765396147803627,
      "median": 0.13636363636363635,
      "min": 0.03125,
      "max": 1.0
    },
    "reference_token_recall": {
      "mean": 0.4790518374906662,
      "median": 0.4065148305084746,
      "min": 0.0,
      "max": 1.0
    },
    "task_counts": {
      "bible_ref": 1024,
      "translate": 1067,
      "usage_example_with_glossary": 365
    },
    "tier_counts": {
      "high_confidence_verse": 1645,
      "none": 408,
      "sentence_candidate": 403
    },
    "db_headword_hits": {
      "rows_with_headword": 365,
      "hits": 274,
      "rate": 0.7506849315068493
    },
    "elder_sentence_pair_exact": {
      "rows": 43,
      "exact": 0,
      "rate": 0.0
    }
  },
  "validation_bible_direct": {
    "metrics": {
      "bleu": 11.980071379735055,
      "chrf": 39.30213095247043,
      "rows": 2153,
      "num_beams": 1,
      "no_repeat_ngram_size": 0,
      "repetition_penalty": 1.0,
      "length_penalty": 1.0
    },
    "rows": 2153,
    "exact": 52,
    "exact_rate": 0.024152345564328843,
    "empty": 1,
    "length_ratio": {
      "mean": 1.000728470567241,
      "median": 0.9047619047619048,
      "min": 0.0,
      "max": 20.0
    },
    "max_token_repeat_share": {
      "mean": 0.1485811698970536,
      "median": 0.1282051282051282,
      "min": 0.0,
      "max": 1.0
    },
    "reference_token_recall": {
      "mean": 0.3743649075183451,
      "median": 0.35,
      "min": 0.0,
      "max": 1.0
    },
    "task_counts": {
      "translate": 2153
    },
    "tier_counts": {
      "high_confidence_verse": 1670,
      "sentence_candidate": 483
    },
    "db_headword_hits": {
      "rows_with_headword": 0,
      "hits": 0,
      "rate": null
    },
    "elder_sentence_pair_exact": {
      "rows": 0,
      "exact": 0,
      "rate": null
    }
  },
  "validation_bible_ref": {
    "metrics": {
      "bleu": 11.934767061577693,
      "chrf": 39.093880392237715,
      "rows": 2153,
      "num_beams": 1,
      "no_repeat_ngram_size": 0,
      "repetition_penalty": 1.0,
      "length_penalty": 1.0
    },
    "rows": 2153,
    "exact": 46,
    "exact_rate": 0.021365536460752437,
    "empty": 0,
    "length_ratio": {
      "mean": 1.0114747554070926,
      "median": 0.9166666666666666,
      "min": 0.16981132075471697,
      "max": 18.666666666666668
    },
    "max_token_repeat_share": {
      "mean": 0.147683715377849,
      "median": 0.12698412698412698,
      "min": 0.03225806451612903,
      "max": 0.9767441860465116
    },
    "reference_token_recall": {
      "mean": 0.3703341408095996,
      "median": 0.34615384615384615,
      "min": 0.0,
      "max": 1.0
    },
    "task_counts": {
      "bible_ref": 2153
    },
    "tier_counts": {
      "high_confidence_verse": 1670,
      "sentence_candidate": 483
    },
    "db_headword_hits": {
      "rows_with_headword": 0,
      "hits": 0,
      "rate": null
    },
    "elder_sentence_pair_exact": {
      "rows": 0,
      "exact": 0,
      "rate": null
    }
  },
  "test_bible_direct": {
    "metrics": {
      "bleu": 11.902008556232007,
      "chrf": 39.41441994560193,
      "rows": 2135,
      "num_beams": 1,
      "no_repeat_ngram_size": 0,
      "repetition_penalty": 1.0,
      "length_penalty": 1.0
    },
    "rows": 2135,
    "exact": 65,
    "exact_rate": 0.03044496487119438,
    "empty": 0,
    "length_ratio": {
      "mean": 1.0160772356245436,
      "median": 0.9090909090909091,
      "min": 0.08791208791208792,
      "max": 14.11111111111111
    },
    "max_token_repeat_share": {
      "mean": 0.1474362650921968,
      "median": 0.125,
      "min": 0.036585365853658534,
      "max": 1.0
    },
    "reference_token_recall": {
      "mean": 0.3760617532217962,
      "median": 0.3333333333333333,
      "min": 0.0,
      "max": 1.0
    },
    "task_counts": {
      "translate": 2135
    },
    "tier_counts": {
      "high_confidence_verse": 1657,
      "sentence_candidate": 478
    },
    "db_headword_hits": {
      "rows_with_headword": 0,
      "hits": 0,
      "rate": null
    },
    "elder_sentence_pair_exact": {
      "rows": 0,
      "exact": 0,
      "rate": null
    }
  },
  "test_bible_ref": {
    "metrics": {
      "bleu": 11.926216963309498,
      "chrf": 39.38908843508707,
      "rows": 2135,
      "num_beams": 1,
      "no_repeat_ngram_size": 0,
      "repetition_penalty": 1.0,
      "length_penalty": 1.0
    },
    "rows": 2135,
    "exact": 67,
    "exact_rate": 0.03138173302107728,
    "empty": 0,
    "length_ratio": {
      "mean": 1.0181939895961498,
      "median": 0.9090909090909091,
      "min": 0.17582417582417584,
      "max": 9.75
    },
    "max_token_repeat_share": {
      "mean": 0.1456170752676329,
      "median": 0.125,
      "min": 0.036585365853658534,
      "max": 0.9431818181818182
    },
    "reference_token_recall": {
      "mean": 0.37688054645959773,
      "median": 0.3333333333333333,
      "min": 0.0,
      "max": 1.0
    },
    "task_counts": {
      "bible_ref": 2135
    },
    "tier_counts": {
      "high_confidence_verse": 1657,
      "sentence_candidate": 478
    },
    "db_headword_hits": {
      "rows_with_headword": 0,
      "hits": 0,
      "rate": null
    },
    "elder_sentence_pair_exact": {
      "rows": 0,
      "exact": 0,
      "rate": null
    }
  },
  "heldout_usage": {
    "metrics": {
      "bleu": 7.835930734519531,
      "chrf": 43.25761176817299,
      "rows": 84,
      "num_beams": 1,
      "no_repeat_ngram_size": 0,
      "repetition_penalty": 1.0,
      "length_penalty": 1.0
    },
    "rows": 84,
    "exact": 3,
    "exact_rate": 0.03571428571428571,
    "empty": 0,
    "length_ratio": {
      "mean": 1.4398809523809524,
      "median": 1.0,
      "min": 0.3333333333333333,
      "max": 34.0
    },
    "max_token_repeat_share": {
      "mean": 0.31207621937663954,
      "median": 0.25,
      "min": 0.1111111111111111,
      "max": 1.0
    },
    "reference_token_recall": {
      "mean": 0.43641345427059713,
      "median": 0.3666666666666667,
      "min": 0.0,
      "max": 1.0
    },
    "task_counts": {
      "usage_example_with_glossary": 84
    },
    "tier_counts": {
      "none": 84
    },
    "db_headword_hits": {
      "rows_with_headword": 84,
      "hits": 59,
      "rate": 0.7023809523809523
    },
    "elder_sentence_pair_exact": {
      "rows": 0,
      "exact": 0,
      "rate": null
    }
  },
  "elder_sentence_pair_all": {
    "metrics": {
      "bleu": 12.64480456889731,
      "chrf": 36.2399979835272,
      "rows": 43,
      "num_beams": 1,
      "no_repeat_ngram_size": 0,
      "repetition_penalty": 1.0,
      "length_penalty": 1.0
    },
    "rows": 43,
    "exact": 0,
    "exact_rate": 0.0,
    "empty": 0,
    "length_ratio": {
      "mean": 0.8335912674879897,
      "median": 0.75,
      "min": 0.47058823529411764,
      "max": 3.6666666666666665
    },
    "max_token_repeat_share": {
      "mean": 0.17017310375003514,
      "median": 0.14285714285714285,
      "min": 0.07142857142857142,
      "max": 0.5
    },
    "reference_token_recall": {
      "mean": 0.2771130400287467,
      "median": 0.23333333333333334,
      "min": 0.0,
      "max": 0.75
    },
    "task_counts": {
      "translate": 43
    },
    "tier_counts": {
      "none": 43
    },
    "db_headword_hits": {
      "rows_with_headword": 0,
      "hits": 0,
      "rate": null
    },
    "elder_sentence_pair_exact": {
      "rows": 43,
      "exact": 0,
      "rate": 0.0
    }
  },
  "test_all": {
    "metrics": {
      "bleu": 11.923665737688653,
      "chrf": 39.39147030888842,
      "rows": 4397,
      "num_beams": 1,
      "no_repeat_ngram_size": 0,
      "repetition_penalty": 1.0,
      "length_penalty": 1.0
    },
    "rows": 4397,
    "exact": 135,
    "exact_rate": 0.030702751876279282,
    "empty": 0,
    "length_ratio": {
      "mean": 1.0234167592331507,
      "median": 0.9090909090909091,
      "min": 0.08791208791208792,
      "max": 34.0
    },
    "max_token_repeat_share": {
      "mean": 0.14992056573962373,
      "median": 0.125,
      "min": 0.036585365853658534,
      "max": 1.0
    },
    "reference_token_recall": {
      "mean": 0.37664462149186767,
      "median": 0.3333333333333333,
      "min": 0.0,
      "max": 1.0
    },
    "task_counts": {
      "bible_ref": 2135,
      "translate": 2178,
      "usage_example_with_glossary": 84
    },
    "tier_counts": {
      "high_confidence_verse": 3314,
      "none": 127,
      "sentence_candidate": 956
    },
    "db_headword_hits": {
      "rows_with_headword": 84,
      "hits": 59,
      "rate": 0.7023809523809523
    },
    "elder_sentence_pair_exact": {
      "rows": 43,
      "exact": 0,
      "rate": 0.0
    }
  }
}
```

## Resource Summary

```json
{
  "samples": 2174,
  "start": "2026-07-02T13:39:22Z",
  "end": "2026-07-02T16:44:14Z",
  "mean_gpu_util_pct": 42.94940202391904,
  "max_gpu_util_pct": 100.0,
  "mean_gpu_mem_used_mib": 17287.45814167433,
  "max_gpu_mem_used_mib": 21914.0,
  "mean_gpu_power_w": 191.50095216191352,
  "max_gpu_power_w": 290.49,
  "mean_python_cpu_pct": 116.66067157313708,
  "max_python_cpu_pct": 537.0,
  "mean_python_rss_mib": 2333.2231830726773,
  "max_python_rss_mib": 7537.6,
  "mean_ram_used_mib": 124068.80036798528,
  "max_ram_used_mib": 205758.0,
  "mean_workspace_used_pct": 67.0,
  "max_workspace_used_pct": 67.0
}
```
