Text-to-Audio
PEFT
Safetensors
audio-generation
lora
flow-matching
midashenglm-gen-wer-lora / history.json
adhikjoshi's picture
Upload folder using huggingface_hub
e9e75f4 verified
Raw
History Blame Contribute Delete
9.1 kB
{
"args": {
"data": "/workspace/train_data_v8",
"out": "/workspace/runs/scene-v8",
"model_dir": "/workspace/models/midashenglm-gen",
"steps": 2500,
"batch_size": 8,
"max_per_capability": 11000,
"patch_budget": 280,
"accum": 2,
"lr": 5e-05,
"lr_min": 1e-06,
"warmup": 150,
"clip": 1.0,
"lambda_stop": 0.01,
"rank": 32,
"dit_rank": 64,
"alpha": 64,
"lora_dropout": 0.05,
"mix": "",
"anchor_weight": 0.5,
"anchor_prob": 0.35,
"short_boost": 2.5,
"onset_boost": 3.0,
"onset_patches": 3,
"scene_llm": false,
"tolerance": 0.02,
"length_tolerance": 0.04,
"eval_every": 250,
"seed": 0,
"hub_repo": "ModelsLab/midashenglm-gen-wer-lora",
"sample_every": 500,
"patience": 4,
"min_delta": 0.001,
"max_steps": 5000,
"extend_steps": 500,
"extra_data": null,
"hub_token": "<redacted>"
},
"trainable": 72876032,
"mix": {
"sfx": 0.22,
"ambience": 0.2,
"speech": 0.2,
"mixed": 0.2,
"music": 0.18
},
"baseline": {
"speech": 0.0956180514767766,
"sfx": 0.2035965397953987,
"music": 0.17061586864292622,
"ambience": 0.20357421226799488,
"mixed": 0.12581421993672848
},
"baseline_stop": {
"speech": 0.9989583333333334,
"sfx": 0.9924806749622928,
"music": 0.9987702546296297,
"ambience": 1.0,
"mixed": 0.9989255536130536
},
"baseline_gen_lengths": {
"sfx-gravel": 10.0,
"sfx-fire": 10.0,
"ambience-rain": 10.0
},
"history": [
{
"step": 250,
"losses": {
"speech": 0.09400662407279015,
"sfx": 0.20097726956009865,
"music": 0.16864320449531078,
"ambience": 0.20278768613934517,
"mixed": 0.12525759171694517
},
"deltas": {
"speech": -0.01685275300111959,
"sfx": -0.012865003687843912,
"music": -0.011562020363673951,
"ambience": -0.0038635842913850543,
"mixed": -0.00442420753443635
},
"mean": 0.15833447519689797,
"stop_ratio": {
"speech": 0.9989583333333334,
"sfx": 0.9924806749622928,
"music": 0.9987702546296297,
"ambience": 1.0,
"mixed": 0.9989255536130536
},
"stop_shrink": {
"speech": 0.0,
"sfx": 0.0,
"music": 0.0,
"ambience": 0.0,
"mixed": 0.0
},
"gen_lengths": {
"sfx-gravel": 10.0,
"sfx-fire": 10.0,
"ambience-rain": 9.8
},
"gen_shrink": {
"sfx-gravel": 0.0,
"sfx-fire": 0.0,
"ambience-rain": 0.019999999999999928
},
"accepted": true,
"stale": 0
},
{
"step": 500,
"losses": {
"speech": 0.09378513786941767,
"sfx": 0.20070120319724083,
"music": 0.16849982738494873,
"ambience": 0.20258150435984135,
"mixed": 0.12504019401967525
},
"deltas": {
"speech": -0.019169116908893554,
"sfx": -0.01422095189371837,
"music": -0.012402370745513988,
"ambience": -0.0048763932184430385,
"mixed": -0.006152133816372089
},
"mean": 0.15812157336622476,
"stop_ratio": {
"speech": 0.9989583333333334,
"sfx": 0.9924806749622928,
"music": 0.9987702546296297,
"ambience": 1.0,
"mixed": 0.9989255536130536
},
"stop_shrink": {
"speech": 0.0,
"sfx": 0.0,
"music": 0.0,
"ambience": 0.0,
"mixed": 0.0
},
"gen_lengths": {
"sfx-gravel": 10.0,
"sfx-fire": 10.0,
"ambience-rain": 10.0
},
"gen_shrink": {
"sfx-gravel": 0.0,
"sfx-fire": 0.0,
"ambience-rain": 0.0
},
"accepted": true,
"stale": 0
},
{
"step": 750,
"losses": {
"speech": 0.09360956773161888,
"sfx": 0.20055881701409817,
"music": 0.16840521432459354,
"ambience": 0.20233918726444244,
"mixed": 0.12497375253587961
},
"deltas": {
"speech": -0.021005277917062894,
"sfx": -0.014920306525608239,
"music": -0.012956909201448576,
"ambience": -0.00606670653317617,
"mixed": -0.006680225822419227
},
"mean": 0.15797730777412652,
"stop_ratio": {
"speech": 0.9989583333333334,
"sfx": 0.9924806749622928,
"music": 0.9987702546296297,
"ambience": 1.0,
"mixed": 0.9989255536130536
},
"stop_shrink": {
"speech": 0.0,
"sfx": 0.0,
"music": 0.0,
"ambience": 0.0,
"mixed": 0.0
},
"gen_lengths": {
"sfx-gravel": 10.0,
"sfx-fire": 10.0,
"ambience-rain": 9.8
},
"gen_shrink": {
"sfx-gravel": 0.0,
"sfx-fire": 0.0,
"ambience-rain": 0.019999999999999928
},
"accepted": true,
"stale": 1
},
{
"step": 1000,
"losses": {
"speech": 0.09348684921860695,
"sfx": 0.2005684357136488,
"music": 0.16837280616164207,
"ambience": 0.202393414452672,
"mixed": 0.1249053431674838
},
"deltas": {
"speech": -0.022288702031198267,
"sfx": -0.014873062601127524,
"music": -0.013146857318286962,
"ambience": -0.0058003310054242875,
"mixed": -0.007223959022292884
},
"mean": 0.15794536974281073,
"stop_ratio": {
"speech": 0.9989583333333334,
"sfx": 0.9924806749622928,
"music": 0.9987702546296297,
"ambience": 1.0,
"mixed": 0.9989255536130536
},
"stop_shrink": {
"speech": 0.0,
"sfx": 0.0,
"music": 0.0,
"ambience": 0.0,
"mixed": 0.0
},
"gen_lengths": {
"sfx-gravel": 10.0,
"sfx-fire": 10.0,
"ambience-rain": 10.0
},
"gen_shrink": {
"sfx-gravel": 0.0,
"sfx-fire": 0.0,
"ambience-rain": 0.0
},
"accepted": true,
"stale": 2
},
{
"step": 1250,
"losses": {
"speech": 0.09341771434992552,
"sfx": 0.20048575475811958,
"music": 0.16836701519787312,
"ambience": 0.20210730656981468,
"mixed": 0.12483794149011374
},
"deltas": {
"speech": -0.023011733588667538,
"sfx": -0.015279164569325519,
"music": -0.013180798849136468,
"ambience": -0.007205754018829721,
"mixed": -0.0077596828649870355
},
"mean": 0.15784314647316933,
"stop_ratio": {
"speech": 0.9989583333333334,
"sfx": 0.9924806749622928,
"music": 0.9984447337962963,
"ambience": 1.0,
"mixed": 0.9989255536130536
},
"stop_shrink": {
"speech": 0.0,
"sfx": 0.0,
"music": 0.0003259216339538286,
"ambience": 0.0,
"mixed": 0.0
},
"gen_lengths": {
"sfx-gravel": 10.0,
"sfx-fire": 10.0,
"ambience-rain": 9.8
},
"gen_shrink": {
"sfx-gravel": 0.0,
"sfx-fire": 0.0,
"ambience-rain": 0.019999999999999928
},
"accepted": true,
"stale": 3
},
{
"step": 1500,
"losses": {
"speech": 0.09332506265491247,
"sfx": 0.20040102116763592,
"music": 0.16829048842191696,
"ambience": 0.2020278088748455,
"mixed": 0.1248040571808815
},
"deltas": {
"speech": -0.023980710613215557,
"sfx": -0.015695348412964598,
"music": -0.013629331430336832,
"ambience": -0.007596263671715041,
"mixed": -0.008029003051920398
},
"mean": 0.15776968766003846,
"stop_ratio": {
"speech": 0.9989583333333334,
"sfx": 0.9924806749622928,
"music": 0.9984447337962963,
"ambience": 1.0,
"mixed": 0.9989255536130536
},
"stop_shrink": {
"speech": 0.0,
"sfx": 0.0,
"music": 0.0003259216339538286,
"ambience": 0.0,
"mixed": 0.0
},
"gen_lengths": {
"sfx-gravel": 10.0,
"sfx-fire": 10.0,
"ambience-rain": 9.8
},
"gen_shrink": {
"sfx-gravel": 0.0,
"sfx-fire": 0.0,
"ambience-rain": 0.019999999999999928
},
"accepted": true,
"stale": 4
}
],
"best": {
"score": 0.15776968766003846,
"step": 1500,
"losses": {
"speech": 0.09332506265491247,
"sfx": 0.20040102116763592,
"music": 0.16829048842191696,
"ambience": 0.2020278088748455,
"mixed": 0.1248040571808815
},
"deltas": {
"speech": -0.023980710613215557,
"sfx": -0.015695348412964598,
"music": -0.013629331430336832,
"ambience": -0.007596263671715041,
"mixed": -0.008029003051920398
},
"stop_ratio": {
"speech": 0.9989583333333334,
"sfx": 0.9924806749622928,
"music": 0.9984447337962963,
"ambience": 1.0,
"mixed": 0.9989255536130536
}
},
"total_steps": 2500
}