Create config_huge.yaml
Browse files- config_huge.yaml +23 -0
config_huge.yaml
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
dataset:
|
| 2 |
+
src_lang: 'lo'
|
| 3 |
+
src_tokenizer: 'BPE'
|
| 4 |
+
src_max_seq_len: 400
|
| 5 |
+
tgt_lang: 'vi'
|
| 6 |
+
tgt_tokenizer: 'WordLevel'
|
| 7 |
+
tgt_max_seq_len: 350
|
| 8 |
+
train_dataset: 'train_clean.dat'
|
| 9 |
+
validate_dataset: 'dev_clean.dat'
|
| 10 |
+
tokenizer_file: "tokenizer_{0}.json"
|
| 11 |
+
bleu_dataset: 'test2023'
|
| 12 |
+
|
| 13 |
+
model:
|
| 14 |
+
d_model: 512
|
| 15 |
+
num_heads: 8
|
| 16 |
+
d_ff: 2048
|
| 17 |
+
dropout_p: 0.15
|
| 18 |
+
num_encoder_layers: 8
|
| 19 |
+
num_decoder_layers: 4
|
| 20 |
+
model_folder: "weights"
|
| 21 |
+
model_basename: "transformer_"
|
| 22 |
+
preload: 'huge'
|
| 23 |
+
|