{
  "model_id": "dragonkue/snowflake-arctic-embed-l-v2.0-ko",
  "variation": "capped_quarter",
  "seed": 20260921,
  "data_seed": 20260921,
  "base_train_rows": 540000,
  "validation_rows": 60000,
  "upsampled_train_population": {
    "base_rows": 540000,
    "extra_score4_rows": 27498,
    "rows": 567498,
    "score4_rows": 56750,
    "score4_fraction": 0.10000035242414951,
    "indices_sha256": "cd146dfd9c4772ef38d09a4bf593d1f213b2bb368824c917a0727d2da1a17bc4"
  },
  "effective_batch_size": 256,
  "train_micro_batches": {
    "512": 32,
    "1024": 16,
    "2048": 8
  },
  "gradient_accumulation": {
    "512": 8,
    "1024": 16,
    "2048": 32
  },
  "context_step_ratio": {
    "512": 16,
    "1024": 4,
    "2048": 1
  },
  "step_counts": {
    "512": 6757,
    "1024": 1689,
    "2048": 422
  },
  "sample_counts": {
    "512": 1729792,
    "1024": 432384,
    "2048": 108032
  },
  "steps_per_epoch": 2217,
  "max_epochs": 4,
  "max_steps": 8868,
  "eval_steps": [
    1109,
    2217,
    3326,
    4434,
    5543,
    6651,
    7760,
    8868
  ],
  "warmup_steps": 300,
  "head_learning_rate": 0.0003,
  "encoder_learning_rate": 3e-05,
  "scheduler": "linear warmup then cosine decay to zero",
  "truncation_marker": null,
  "split_manifest": {
    "created_at": "2026-10-03T18:28:15.232161+00:00",
    "seed": 20260921,
    "method": "stratified by public/private provenance, content score, integrity label; cluster-safe",
    "base_rows": 600000,
    "train": {
      "rows": 540000,
      "public_rows": 449998,
      "private_rows": 90002,
      "content_counts": {
        "0": 60618,
        "1": 251387,
        "2": 128263,
        "3": 70480,
        "4": 29252
      },
      "integrity_counts": {
        "0": 528496,
        "1": 9624,
        "2": 1880
      },
      "indices_sha256": "d0384fb70fbea90859eb71fd10f0a34b654a6c5557089429979533631847d645",
      "ids_sha256": "3241fac0b92bf334ee406d1c8fb868bad11be577433c73b35043e1d92f03b728",
      "clusters_sha256": "2c0728e27091ce4e2699cd0cf25fba53792209b75cb040dc6d4d568e70a9f01d"
    },
    "validation": {
      "rows": 60000,
      "public_rows": 50002,
      "private_rows": 9998,
      "content_counts": {
        "0": 6735,
        "1": 27932,
        "2": 14252,
        "3": 7830,
        "4": 3251
      },
      "integrity_counts": {
        "0": 58720,
        "1": 1070,
        "2": 210
      },
      "indices_sha256": "80d7239d50b44a23d2ccab4e6a5af50d789be7e4977d8f6172783cc8bde72e9d",
      "ids_sha256": "05a6505df655a1edee4bbfcf66e055d1b1c63f3a8a7cb61677cc430788e7fd4f",
      "clusters_sha256": "36222da42a8344e1381e320a00190f58af507d39afdc9a88096a07df0bbbe14b"
    },
    "overlap": {
      "indices": 0,
      "clusters": 0
    }
  },
  "schedule_sha256": "06270c088964c90c980ab8856de5b8f3b1c5bb4c478f53100e944bf1078432f3",
  "population_indices_sha256": "ad8477c94140db4b9f66b7cd0320da8e760c8577f8c67bde41849a56d3b59f7e",
  "checkpoint_selection": {
    "objectives": [
      "validation content macro F1",
      "validation mean content F1 classes 2 and 3"
    ],
    "dominated_definition": "both objectives strictly lower than an earlier checkpoint",
    "minimum_epochs": 2,
    "patience": "two consecutive half-epoch dominated checkpoints"
  }
}