RootAccess4Life commited on
Commit
d567519
·
verified ·
1 Parent(s): ed4d7f8

Upload phase2-hubert-a085/config.yaml with huggingface_hub

Browse files
Files changed (1) hide show
  1. phase2-hubert-a085/config.yaml +413 -0
phase2-hubert-a085/config.yaml ADDED
@@ -0,0 +1,413 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ accumulate_grad_batches: 1
2
+ add_word_pos: true
3
+ amp: true
4
+ ast_context_model: MIT/ast-finetuned-audioset-10-10-0.4593
5
+ ast_context_scale: 0.25
6
+ ast_context_valid_prefix: false
7
+ audio_num_mel_bins: 100
8
+ audio_sample_rate: 24000
9
+ binarizer_cls: data_gen.tts.base_binarizer.BaseBinarizer
10
+ binary_data_dir: /mnt/DATA/ResearchDrive/audiopaint/data/binary/libritts_bigvgan
11
+ cgpc_gt_view: utt
12
+ cgpc_tau: 0.1
13
+ clip_grad_norm: 1
14
+ clip_grad_value: 0
15
+ cond_dim: 0
16
+ dec_dilations:
17
+ - 1
18
+ - 1
19
+ - 1
20
+ - 1
21
+ dec_ffn_kernel_size: 9
22
+ dec_inp_add_noise: false
23
+ dec_kernel_size: 5
24
+ dec_layers: 4
25
+ dec_post_net_kernel: 3
26
+ decoder_rnn_dim: 0
27
+ decoder_type: conv
28
+ detach_postflow_input: true
29
+ diff_decoder_type: wavenet
30
+ diff_loss_type: l1
31
+ dilation_cycle_length: 1
32
+ dropout: 0.0
33
+ ds_name: libritts
34
+ ds_workers: 2
35
+ dur_level: word
36
+ dur_predictor_kernel: 5
37
+ dur_predictor_layers: 3
38
+ enc_dec_norm: ln
39
+ enc_dilations:
40
+ - 1
41
+ - 1
42
+ - 1
43
+ - 1
44
+ enc_ffn_kernel_size: 5
45
+ enc_kernel_size: 5
46
+ enc_layers: 4
47
+ enc_post_net_kernel: 3
48
+ enc_pre_ln: true
49
+ enc_prenet: true
50
+ encoder_K: 8
51
+ encoder_type: conv
52
+ endless_ds: true
53
+ eval_max_batches: -1
54
+ f0_max: 600
55
+ f0_min: 80
56
+ ffn_hidden_size: 768
57
+ fft_size: 1024
58
+ fmax: null
59
+ fmin: 0
60
+ frames_multiple: 1
61
+ gen_dir_name: ''
62
+ gst_seed: 1234
63
+ gst_trainable: false
64
+ hidden_size: 192
65
+ hlac_frame_metric: norm
66
+ hlac_levels: fpw
67
+ hlac_w_frame: 1.0
68
+ hlac_w_phone: 1.0
69
+ hlac_w_word: 1.0
70
+ hop_size: 256
71
+ hubert_feature_context_frames: 12
72
+ hubert_feature_crop_frames: 64
73
+ hubert_feature_items_per_batch: 1
74
+ hubert_feature_layer: conv
75
+ infer_mask_ratio: 0.3
76
+ keep_bins: 100
77
+ kl_min: 0.0
78
+ kl_start_steps: 10000
79
+ lambda_cgpc: 0.0
80
+ lambda_energy: 0.1
81
+ lambda_f0: 1.0
82
+ lambda_hlac: 0.0
83
+ lambda_hubert_feature: 0.0
84
+ lambda_kl: 1.0
85
+ lambda_mel_adv: 0.05
86
+ lambda_ph_dur: 0.1
87
+ lambda_phone_mel_contrast: 0.0
88
+ lambda_sent_dur: 0.0
89
+ lambda_uv: 1.0
90
+ lambda_word_dur: 1.0
91
+ latent_size: 16
92
+ layers_in_block: 2
93
+ load_ckpt: /mnt/DATA/ResearchDrive/audiopaint/parents/best/model_ckpt_steps_5000.ckpt
94
+ load_ckpt_min_match: 0.9
95
+ loud_norm: false
96
+ lr: 2.0e-05
97
+ mask_type: continuous_
98
+ max_beta: 0.06
99
+ max_frames: 1548
100
+ max_input_tokens: 1550
101
+ max_sentences: 16
102
+ max_tokens: 40000
103
+ max_updates: 1500
104
+ max_valid_sentences: 1
105
+ max_valid_tokens: 60000
106
+ mel_disc_hidden_size: 128
107
+ mel_losses: l1:0.5|ssim:0.5
108
+ mel_type: bigvgan
109
+ mel_vmax: 1.5
110
+ mel_vmin: -6
111
+ min_frames: 0
112
+ noise_scale: 0.8
113
+ num_ckpt_keep: 3
114
+ num_heads: 2
115
+ num_sanity_val_steps: 5
116
+ num_spk: 1261
117
+ num_valid_plots: 10
118
+ optimizer_adam_beta1: 0.9
119
+ optimizer_adam_beta2: 0.98
120
+ out_wav_norm: false
121
+ phone_mel_prototypes: assets/agent-1_phone_mel_prototypes.npz
122
+ phone_mel_temperature: 0.1
123
+ pitch_extractor: parselmouth
124
+ pitch_type: frame
125
+ posterior_start_steps: 0
126
+ predictor_dropout: 0.2
127
+ predictor_grad: 0.1
128
+ predictor_hidden: -1
129
+ predictor_kernel: 5
130
+ predictor_layers: 5
131
+ print_nan_grads: false
132
+ prior_flow_hidden: 64
133
+ prior_flow_kernel_size: 3
134
+ prior_flow_n_blocks: 4
135
+ processed_data_dir: /mnt/DATA/ResearchDrive/audiopaint/data/processed/libritts
136
+ profile_infer: false
137
+ raw_data_dir: data/raw/libritts
138
+ residual_channels: 256
139
+ residual_layers: 20
140
+ resume_from_checkpoint: 0
141
+ rope_half: true
142
+ rope_reset_at_nt: true
143
+ save_best: false
144
+ save_codes:
145
+ - tasks
146
+ - modules
147
+ save_gt: true
148
+ schedule_type: vpsde
149
+ scheduler: warmup
150
+ seed: 1234
151
+ share_wn_layers: 4
152
+ sigmoid_scale: false
153
+ sort_by_len: true
154
+ spec_max:
155
+ - 1.0604
156
+ - 1.4571
157
+ - 1.2852
158
+ - 1.3722
159
+ - 1.1957
160
+ - 1.6533
161
+ - 1.5299
162
+ - 1.1385
163
+ - 1.2986
164
+ - 1.3136
165
+ - 1.5607
166
+ - 1.3109
167
+ - 1.7511
168
+ - 1.7297
169
+ - 1.5513
170
+ - 1.6957
171
+ - 1.3587
172
+ - 1.8329
173
+ - 1.4253
174
+ - 1.2918
175
+ - 1.4762
176
+ - 1.3411
177
+ - 1.4748
178
+ - 1.3285
179
+ - 1.4044
180
+ - 1.1602
181
+ - 1.4003
182
+ - 1.2423
183
+ - 1.2876
184
+ - 1.0298
185
+ - 1.2514
186
+ - 1.1693
187
+ - 1.2147
188
+ - 1.1453
189
+ - 1.055
190
+ - 1.2303
191
+ - 1.4054
192
+ - 1.0691
193
+ - 1.1672
194
+ - 1.21
195
+ - 0.6927
196
+ - 0.6037
197
+ - 0.9786
198
+ - 0.361
199
+ - 0.7867
200
+ - 0.3277
201
+ - 0.4398
202
+ - 0.5989
203
+ - 0.4387
204
+ - 0.3292
205
+ - 0.1331
206
+ - 0.2449
207
+ - 0.2704
208
+ - -0.1298
209
+ - 0.0984
210
+ - -0.1073
211
+ - -0.1773
212
+ - -0.1167
213
+ - -0.1131
214
+ - -0.1729
215
+ - -0.3771
216
+ - -0.3133
217
+ - -0.5095
218
+ - -0.6342
219
+ - -0.1225
220
+ - -0.0924
221
+ - -0.2996
222
+ - -0.1631
223
+ - -0.4548
224
+ - -0.6299
225
+ - -0.4956
226
+ - -0.2327
227
+ - -0.2719
228
+ - -0.8649
229
+ - -0.4402
230
+ - -0.6376
231
+ - -0.7074
232
+ - -0.552
233
+ - -0.7659
234
+ - -0.7273
235
+ - -0.8326
236
+ - -0.7609
237
+ - -0.9281
238
+ - -0.6764
239
+ - -0.3611
240
+ - -1.0908
241
+ - -1.167
242
+ - -0.9377
243
+ - -0.699
244
+ - -0.7402
245
+ - -1.1614
246
+ - -1.1782
247
+ - -0.6824
248
+ - -0.8936
249
+ - -0.7902
250
+ - -1.1805
251
+ - -1.3148
252
+ - -1.3482
253
+ - -1.61
254
+ - -2.2738
255
+ spec_min:
256
+ - -11.5129
257
+ - -11.5129
258
+ - -11.5129
259
+ - -11.5129
260
+ - -11.5129
261
+ - -11.5129
262
+ - -11.5129
263
+ - -11.5129
264
+ - -11.5129
265
+ - -11.5129
266
+ - -11.5129
267
+ - -11.5129
268
+ - -11.5129
269
+ - -11.5129
270
+ - -11.5129
271
+ - -11.5129
272
+ - -11.5129
273
+ - -11.5129
274
+ - -11.5129
275
+ - -11.5129
276
+ - -11.5129
277
+ - -11.5129
278
+ - -11.5129
279
+ - -11.5129
280
+ - -11.5129
281
+ - -11.5129
282
+ - -11.5129
283
+ - -11.5129
284
+ - -11.5129
285
+ - -11.5129
286
+ - -11.5129
287
+ - -11.5129
288
+ - -11.5129
289
+ - -11.5129
290
+ - -11.5129
291
+ - -11.5129
292
+ - -11.5129
293
+ - -11.5129
294
+ - -11.5129
295
+ - -11.5129
296
+ - -11.5129
297
+ - -11.5129
298
+ - -11.5129
299
+ - -11.5129
300
+ - -11.5129
301
+ - -11.5129
302
+ - -11.5129
303
+ - -11.5129
304
+ - -11.5129
305
+ - -11.5129
306
+ - -11.5129
307
+ - -11.5129
308
+ - -11.5129
309
+ - -11.5129
310
+ - -11.5129
311
+ - -11.5129
312
+ - -11.5129
313
+ - -11.5129
314
+ - -11.5129
315
+ - -11.5129
316
+ - -11.5129
317
+ - -11.5129
318
+ - -11.5129
319
+ - -11.5129
320
+ - -11.5129
321
+ - -11.5129
322
+ - -11.5129
323
+ - -11.5129
324
+ - -11.5129
325
+ - -11.5129
326
+ - -11.5129
327
+ - -11.5129
328
+ - -11.5129
329
+ - -11.5129
330
+ - -11.5129
331
+ - -11.5129
332
+ - -11.5129
333
+ - -11.5129
334
+ - -11.5129
335
+ - -11.5129
336
+ - -11.5129
337
+ - -11.5129
338
+ - -11.5129
339
+ - -11.5129
340
+ - -11.5129
341
+ - -11.5129
342
+ - -11.5129
343
+ - -11.5129
344
+ - -11.5129
345
+ - -11.5129
346
+ - -11.5129
347
+ - -11.5129
348
+ - -11.5129
349
+ - -11.5129
350
+ - -11.5129
351
+ - -11.5129
352
+ - -11.5129
353
+ - -11.5129
354
+ - -11.5129
355
+ - -11.5129
356
+ task_cls: tasks.speech_editing.spec_denoiser.SpeechDenoiserTask
357
+ tb_log_interval: 100
358
+ test_ids: []
359
+ test_set_name: test
360
+ text_encoder_postnet: true
361
+ timescale: 1
362
+ timesteps: 8
363
+ train_set_name: train
364
+ train_sets: ''
365
+ trainable_only: []
366
+ training_mask_ratio: 0.8
367
+ trunk_adaptation: audio_lora
368
+ trunk_donor: Qwen/Qwen2.5-0.5B
369
+ trunk_freeze_tok_emb: true
370
+ trunk_grad_ckpt: true
371
+ trunk_inject: add
372
+ trunk_layer_pick: ends
373
+ trunk_layers: 14
374
+ trunk_lora_alpha: 16.0
375
+ trunk_lora_rank: 8
376
+ trunk_max_text: 48
377
+ trunk_pretrained_lr_scale: 0.05
378
+ trunk_use_ph: true
379
+ trunk_use_text: true
380
+ trunk_warm_start: true
381
+ two_sided_mask_min_ctx_frac: 0.1
382
+ two_sided_mask_prob: 0.0
383
+ two_sided_mask_ratio: 0.3
384
+ two_stage: true
385
+ use_ast_context: false
386
+ use_cond_proj: false
387
+ use_conn_loss: true
388
+ use_context_mask: false
389
+ use_dualffn_trunk: true
390
+ use_energy_embed: false
391
+ use_gt_dur: false
392
+ use_gt_f0: false
393
+ use_latent_cond: false
394
+ use_mask_phone_gate: false
395
+ use_perceptual_loss: true
396
+ use_pitch_embed: true
397
+ use_post_flow: true
398
+ use_prior_flow: true
399
+ use_spk_embed: true
400
+ use_spk_id: false
401
+ use_txt_cond: true
402
+ use_uv: true
403
+ val_check_interval: 1500
404
+ valid_infer_interval: 2000
405
+ valid_monitor_key: val_loss
406
+ valid_monitor_mode: min
407
+ valid_set_name: valid
408
+ vocoder: BigVGAN
409
+ vocoder_ckpt: pretrained/hifigan_hifitts
410
+ warmup_updates: 1
411
+ weight_decay: 0
412
+ win_size: 1024
413
+ work_dir: /mnt/DATA/ResearchDrive/audiopaint/checkpoints/agent-6_hubert_dn_1500