Upload config.yaml
Browse files- config.yaml +54 -0
config.yaml
ADDED
|
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
hydra:
|
| 2 |
+
run:
|
| 3 |
+
dir: ckpts/${datasets.name}_{model.tokenizer}
|
| 4 |
+
|
| 5 |
+
datasets:
|
| 6 |
+
name: ${datasets.name}
|
| 7 |
+
batch_size_per_gpu: 8
|
| 8 |
+
batch_size_type: sample
|
| 9 |
+
max_samples: 1
|
| 10 |
+
num_workers: 16
|
| 11 |
+
|
| 12 |
+
optim:
|
| 13 |
+
epochs: 250
|
| 14 |
+
learning_rate: 7.5e-5
|
| 15 |
+
num_warmup_updates: 20000
|
| 16 |
+
grad_accumulation_steps: 1
|
| 17 |
+
max_grad_norm: 1.0
|
| 18 |
+
bnb_optimizer: False
|
| 19 |
+
|
| 20 |
+
model:
|
| 21 |
+
name: SilmaTTS_V1_Small
|
| 22 |
+
tokenizer: char
|
| 23 |
+
tokenizer_path: null
|
| 24 |
+
backbone: DiT
|
| 25 |
+
arch:
|
| 26 |
+
dim: 768
|
| 27 |
+
depth: 18
|
| 28 |
+
heads: 12
|
| 29 |
+
ff_mult: 2
|
| 30 |
+
text_dim: 512
|
| 31 |
+
text_mask_padding: True
|
| 32 |
+
conv_layers: 4
|
| 33 |
+
pe_attn_head: null
|
| 34 |
+
attn_backend: torch
|
| 35 |
+
attn_mask_enabled: False
|
| 36 |
+
checkpoint_activations: False
|
| 37 |
+
mel_spec:
|
| 38 |
+
target_sample_rate: 24000
|
| 39 |
+
n_mel_channels: 100
|
| 40 |
+
hop_length: 256
|
| 41 |
+
win_length: 1024
|
| 42 |
+
n_fft: 1024
|
| 43 |
+
mel_spec_type: vocos
|
| 44 |
+
vocoder:
|
| 45 |
+
is_local: False
|
| 46 |
+
local_path: null
|
| 47 |
+
|
| 48 |
+
ckpts:
|
| 49 |
+
logger: wandb
|
| 50 |
+
log_samples: True
|
| 51 |
+
save_per_updates: 1000
|
| 52 |
+
keep_last_n_checkpoints: 100
|
| 53 |
+
last_per_updates: 1000
|
| 54 |
+
save_dir: ckpts/${datasets.name}_{model.tokenizer}
|