ChristophSchuhmann's picture
Release merged NotebookLM full FT + rank-64 final 652
9771537 verified
Raw History Blame Contribute Delete
5.42 kB
{
"status": "READY",
"pilot_only": false,
"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"
},
"base_weights_sha256": "b66725ff44243b7b59c84fba9c201d8f993551e2b3c79211f390042a41f07c06",
"database_sha256": "0e36a157e1810673d15a50fe1d1fb31cf36bafdb45fe09d5ba859dc3abeb6c14",
"presentations": 718464,
"nonpadding_presentations": 718281,
"unique_train_targets": 239427,
"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
}
],
"canary_keys": [
1,
372,
4176,
16,
23624,
222170,
66970
],
"validation_keys": [
4470,
4472,
4473,
4474,
194214,
194215,
194217,
194219,
294871,
294872,
294873,
294874,
19459,
19460,
19461,
19464,
75081,
75084,
75089,
75090
],
"evaluation_tasks": 50,
"evaluation_sha256": "1fd7b56db53a096031dedd274aa60ea7c6546181a66d6f0ab5037d1a40a58a0a",
"episode_split_verified": true,
"statistics": [
{
"language": "de",
"split": "test",
"windows": 93,
"codec_hours": 0.1506222222222222
},
{
"language": "de",
"split": "train",
"windows": 4997,
"codec_hours": 9.151177777777777
},
{
"language": "de",
"split": "validation",
"windows": 46,
"codec_hours": 0.0971777777777778
},
{
"language": "en",
"split": "test",
"windows": 161,
"codec_hours": 0.2633111111111111
},
{
"language": "en",
"split": "train",
"windows": 207561,
"codec_hours": 310.92517777777783
},
{
"language": "en",
"split": "validation",
"windows": 3902,
"codec_hours": 5.888955555555556
},
{
"language": "es",
"split": "test",
"windows": 70,
"codec_hours": 0.10746666666666667
},
{
"language": "es",
"split": "train",
"windows": 25380,
"codec_hours": 33.64906666666667
},
{
"language": "es",
"split": "validation",
"windows": 368,
"codec_hours": 0.48633333333333334
},
{
"language": "fr",
"split": "test",
"windows": 144,
"codec_hours": 0.1645777777777778
},
{
"language": "fr",
"split": "train",
"windows": 820,
"codec_hours": 0.6252222222222222
},
{
"language": "fr",
"split": "validation",
"windows": 149,
"codec_hours": 0.1634888888888889
},
{
"language": "nl",
"split": "test",
"windows": 105,
"codec_hours": 0.22775555555555557
},
{
"language": "nl",
"split": "train",
"windows": 669,
"codec_hours": 1.0362
},
{
"language": "nl",
"split": "validation",
"windows": 74,
"codec_hours": 0.14373333333333335
}
],
"rejections": [
{
"reason": "boundary_contains_other_words",
"candidates": 677661
},
{
"reason": "not_exactly_two_diarized_speakers",
"candidates": 985
},
{
"reason": "target_length",
"candidates": 27339
},
{
"reason": "two_clean_past_references_unavailable",
"candidates": 95293
}
],
"annotation_sha256": "ffe152f39d0d651625fa7747b8d132b587642773b48196e44d5923cfd432e1fd",
"time": 1793631377.1572342
}