{
  "kind": "large13-notebooklm-multispeaker-v1",
  "model": "large",
  "base": "/e/scratch/laionize/schuhmann1/hvs_large_1_3_20261022/release_1_3/model",
  "base_weights_sha256": "b66725ff44243b7b59c84fba9c201d8f993551e2b3c79211f390042a41f07c06",
  "database_sha256": "0e36a157e1810673d15a50fe1d1fb31cf36bafdb45fe09d5ba859dc3abeb6c14",
  "presentations": 718464,
  "actual_presentations": 718281,
  "phases": [
    {
      "name": "Globally shuffled epoch 1",
      "end_position": 239488
    },
    {
      "name": "Globally shuffled epoch 2",
      "end_position": 478976
    },
    {
      "name": "Globally shuffled epoch 3",
      "end_position": 718464
    }
  ],
  "selection_contract": {
    "kind": "large13-notebooklm-multispeaker-v1",
    "parent": "laion/Humaneness-Voice-Large-1.3-Alpha",
    "base_step": 9799,
    "base_weight_files": {
      "model-00001-of-00003.safetensors": "3a9831cfd1944316e8b7d5fcaea845bf8b49b832824950709ce0eabf56b4702a",
      "model-00002-of-00003.safetensors": "1fdaae7bd4f09b867152ef91761106e7bfef986b5f9979f69463466e864018a5",
      "model-00003-of-00003.safetensors": "c149c250be927df24cdaa59c5b8ff50c3f0471f664f6483bb58236f07afa31df"
    },
    "languages": [
      "en",
      "de",
      "fr",
      "es",
      "nl"
    ],
    "epochs": 3,
    "world": 64,
    "accumulation": 2,
    "peaks": [
      2e-05,
      4e-05
    ],
    "warmup": 0.05,
    "cosine_floor": 0.1,
    "prefix": "plain-text speaker-labeled references plus strictly past audio context; native audio placeholders; no new special tokens",
    "context_max_frames": 750,
    "reference_min_frames": 63,
    "reference_max_frames": 187,
    "target_max_frames": 2250,
    "max_context_rows": 32768,
    "loss": "only final assistant audio, all 12 trained codebooks",
    "split": "episode-disjoint, stratified by publisher language; duplicate audio audited",
    "source_filter": "original beginning and middle DNSMOS BAK each >=4.0",
    "eval": {
      "per_language": {
        "two_turn": 5,
        "four_turn": 5
      },
      "seeds": "fixed before training",
      "heldout_episodes": true
    },
    "checkpoints_active_seconds": 900,
    "maximum_training_allocations": 3,
    "no_lora": true,
    "distribution": "FSDP1 root FULL_SHARD; FP32 optimizer parameters/reduction; BF16 compute; full CPU state checkpoints",
    "serialization": "Explicit ChatML with byte-identical train/inference prefix; no template fallback or direct-generation shortcut",
    "published_scope": "Only the user-requested en/de/fr/es/nl source episodes; other languages remain in original journals and archives"
  },
  "world": 64,
  "accumulation": 2,
  "effective_batch": 128,
  "seed": 20261102,
  "peak_semantic": 2e-05,
  "peak_talker": 4e-05,
  "warmup_fraction": 0.05,
  "cosine_floor": 0.1,
  "lr_restarts": 0,
  "total_steps": 5613,
  "checkpoints_seconds": 900,
  "keep_all_checkpoints": true,
  "optimizer_rng_scheduler_sampler_saved": true,
  "pilot": false,
  "distribution": "FSDP1 root FULL_SHARD, original FP32 optimizer parameters, BF16 compute, FP32 reduction, full CPU checkpoint",
  "training_code_sha256": "118513d7a48bc36782c5b8ad7333013ad9fcbf2e40814e7f41ab8cb2a60f75e1",
  "parent_train_fullft_sha256": "4e5f84b948afd3eabd28f762349d1a2c7429232ffc5bd6524f891c82c6a613d3",
  "parent_attention_sha256": "98e46fbe74b8b4e8d48ff12e91a40ce79aab59ea2a970202088a5f91d53bb2cf",
  "snapshot_sha256": "600c0ad3c0d72f1adbfa49cb097f1c983cfc17acf519ad7dc729c341bd5ba45d",
  "variant": "podcast",
  "contract_sha256": "795b57ea6cfca9f0b264e20f2bd5da797d0e9a00e998fb2fc3c5c25a681e818d"
}