Spaces:
Running
on
Zero
Running
on
Zero
brianvlezbr
commited on
Upload 2 files
Browse files- apollo_model_uni.ckpt +3 -0
- config_apollo_uni.yaml +100 -0
apollo_model_uni.ckpt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:11433256eca90143d32f1108d156ac756fd0201fbc69e5f736def4062fb52186
|
3 |
+
size 146867952
|
config_apollo_uni.yaml
ADDED
@@ -0,0 +1,100 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
exp:
|
2 |
+
dir: ./Exps
|
3 |
+
name: Apollo
|
4 |
+
datas:
|
5 |
+
_target_: look2hear.datas.MusdbMoisesdbDataModule
|
6 |
+
train_dir: ./dataset/restoration/train
|
7 |
+
eval_dir: ./dataset/restoration/test
|
8 |
+
codec_type: mp3
|
9 |
+
codec_options:
|
10 |
+
bitrate: random
|
11 |
+
compression: random
|
12 |
+
complexity: random
|
13 |
+
vbr: random
|
14 |
+
sr: 44100
|
15 |
+
segments: 5.4
|
16 |
+
num_stems: 8
|
17 |
+
snr_range:
|
18 |
+
- -10
|
19 |
+
- 10
|
20 |
+
num_samples: 3000
|
21 |
+
batch_size: 1
|
22 |
+
num_workers: 8
|
23 |
+
model:
|
24 |
+
_target_: look2hear.models.apollo.Apollo
|
25 |
+
sr: 44100
|
26 |
+
win: 20
|
27 |
+
feature_dim: 384
|
28 |
+
layer: 6
|
29 |
+
discriminator:
|
30 |
+
_target_: look2hear.discriminators.frequencydis.MultiFrequencyDiscriminator
|
31 |
+
nch: 2
|
32 |
+
window:
|
33 |
+
- 32
|
34 |
+
- 64
|
35 |
+
- 128
|
36 |
+
- 256
|
37 |
+
- 512
|
38 |
+
- 1024
|
39 |
+
- 2048
|
40 |
+
optimizer_g:
|
41 |
+
_target_: bitsandbytes.optim.AdamW8bit
|
42 |
+
lr: 0.001
|
43 |
+
weight_decay: 0.01
|
44 |
+
optimizer_d:
|
45 |
+
_target_: bitsandbytes.optim.AdamW8bit
|
46 |
+
lr: 0.0001
|
47 |
+
weight_decay: 0.01
|
48 |
+
betas:
|
49 |
+
- 0.5
|
50 |
+
- 0.99
|
51 |
+
scheduler_g:
|
52 |
+
_target_: torch.optim.lr_scheduler.StepLR
|
53 |
+
step_size: 4
|
54 |
+
gamma: 0.98
|
55 |
+
scheduler_d:
|
56 |
+
_target_: torch.optim.lr_scheduler.StepLR
|
57 |
+
step_size: 4
|
58 |
+
gamma: 0.98
|
59 |
+
loss_g:
|
60 |
+
_target_: look2hear.losses.gan_losses.MultiFrequencyGenLoss
|
61 |
+
eps: 2.0e-08
|
62 |
+
loss_d:
|
63 |
+
_target_: look2hear.losses.gan_losses.MultiFrequencyDisLoss
|
64 |
+
eps: 2.0e-08
|
65 |
+
metrics:
|
66 |
+
_target_: look2hear.losses.MultiSrcNegSDR
|
67 |
+
sdr_type: sisdr
|
68 |
+
system:
|
69 |
+
_target_: look2hear.system.audio_litmodule.AudioLightningModule
|
70 |
+
early_stopping:
|
71 |
+
_target_: pytorch_lightning.callbacks.EarlyStopping
|
72 |
+
monitor: val_loss
|
73 |
+
patience: 50
|
74 |
+
mode: min
|
75 |
+
verbose: true
|
76 |
+
checkpoint:
|
77 |
+
_target_: pytorch_lightning.callbacks.ModelCheckpoint
|
78 |
+
dirpath: ${exp.dir}/${exp.name}/checkpoints
|
79 |
+
monitor: val_loss
|
80 |
+
mode: min
|
81 |
+
verbose: true
|
82 |
+
save_top_k: 5
|
83 |
+
save_last: true
|
84 |
+
filename: '{epoch}-{val_loss:.4f}'
|
85 |
+
logger:
|
86 |
+
_target_: pytorch_lightning.loggers.WandbLogger
|
87 |
+
name: ${exp.name}
|
88 |
+
save_dir: ${exp.dir}/${exp.name}/logs
|
89 |
+
offline: true
|
90 |
+
project: Audio-Restoration
|
91 |
+
trainer:
|
92 |
+
_target_: pytorch_lightning.Trainer
|
93 |
+
devices:
|
94 |
+
- 0
|
95 |
+
max_epochs: 500
|
96 |
+
sync_batchnorm: true
|
97 |
+
default_root_dir: ${exp.dir}/${exp.name}/
|
98 |
+
accelerator: cuda
|
99 |
+
limit_train_batches: 1.0
|
100 |
+
fast_dev_run: false
|