Create config.yml
Browse files- config.yml +30 -0
config.yml
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
log_dir: "Checkpoint_all_phonemes"
|
2 |
+
mixed_precision: "fp16"
|
3 |
+
data_folder: "wikipedia_20220301.en.processed"
|
4 |
+
batch_size: 32
|
5 |
+
save_interval: 20000
|
6 |
+
log_interval: 10
|
7 |
+
num_process: 1 # number of GPUs
|
8 |
+
num_steps: 2000000
|
9 |
+
|
10 |
+
dataset_params:
|
11 |
+
tokenizer: "bert-base-multilingual-cased"
|
12 |
+
token_separator: " " # token used for phoneme separator (space)
|
13 |
+
token_mask: "M" # token used for phoneme mask (M)
|
14 |
+
word_separator: 102 # token used for word separator (<formula>)
|
15 |
+
token_maps: "token_maps.pkl" # token map path
|
16 |
+
|
17 |
+
max_mel_length: 512 # max phoneme length
|
18 |
+
|
19 |
+
word_mask_prob: 0.15 # probability to mask the entire word
|
20 |
+
phoneme_mask_prob: 0.1 # probability to mask each phoneme
|
21 |
+
replace_prob: 0.2 # probablity to replace phonemes
|
22 |
+
|
23 |
+
model_params:
|
24 |
+
vocab_size: 178
|
25 |
+
hidden_size: 768
|
26 |
+
num_attention_heads: 12
|
27 |
+
intermediate_size: 2048
|
28 |
+
max_position_embeddings: 512
|
29 |
+
num_hidden_layers: 12
|
30 |
+
dropout: 0.1
|