daotthuanSE commited on
Commit
cda9bdf
·
verified ·
1 Parent(s): c4ebbb7

Upload nlp_csanmt_translation_zh2en\configuration.json with huggingface_hub

Browse files
nlp_csanmt_translation_zh2en//configuration.json ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "framework": "tensorflow",
3
+ "task": "translation",
4
+ "pipeline": {
5
+ "type": "csanmt-translation"
6
+ },
7
+ "model": {
8
+ "type": "csanmt-translation",
9
+ "hidden_size": 1024,
10
+ "filter_size": 4096,
11
+ "num_heads": 16,
12
+ "num_encoder_layers": 24,
13
+ "num_decoder_layers": 6,
14
+ "attention_dropout": 0.0,
15
+ "residual_dropout": 0.0,
16
+ "relu_dropout": 0.0,
17
+ "layer_preproc": "layer_norm",
18
+ "layer_postproc": "none",
19
+ "shared_embedding_and_softmax_weights": true,
20
+ "shared_source_target_embedding": true,
21
+ "initializer_scale": 0.1,
22
+ "position_info_type": "absolute",
23
+ "max_relative_dis": 16,
24
+ "num_semantic_encoder_layers": 4,
25
+ "src_vocab_size": 50000,
26
+ "trg_vocab_size": 50000,
27
+ "seed": 1234,
28
+ "beam_size": 4,
29
+ "lp_rate": 0.6,
30
+ "max_decoded_trg_len": 100
31
+ },
32
+ "dataset": {
33
+ "train_src": "train.zh",
34
+ "train_trg": "train.en",
35
+ "src_vocab": {
36
+ "file": "src_vocab.txt"
37
+ },
38
+ "trg_vocab": {
39
+ "file": "trg_vocab.txt"
40
+ }
41
+ },
42
+ "preprocessor": {
43
+ "src_lang": "zh",
44
+ "tgt_lang": "en",
45
+ "src_bpe": {
46
+ "file": "bpe.zh"
47
+ }
48
+ },
49
+ "train": {
50
+ "num_gpus": 0,
51
+ "warmup_steps": 4000,
52
+ "update_cycle": 1,
53
+ "keep_checkpoint_max": 1,
54
+ "confidence": 0.9,
55
+ "optimizer": "adam",
56
+ "adam_beta1": 0.9,
57
+ "adam_beta2": 0.98,
58
+ "adam_epsilon": 1e-9,
59
+ "gradient_clip_norm": 0.0,
60
+ "learning_rate_decay": "linear_warmup_rsqrt_decay",
61
+ "initializer": "uniform_unit_scaling",
62
+ "initializer_scale": 0.1,
63
+ "learning_rate": 1.0,
64
+ "train_batch_size_words": 1024,
65
+ "scale_l1": 0.0,
66
+ "scale_l2": 0.0,
67
+ "train_max_len": 100,
68
+ "num_of_epochs": 2,
69
+ "save_checkpoints_steps": 1000,
70
+ "num_of_samples": 4,
71
+ "eta": 0.6
72
+ },
73
+ "evaluation": {
74
+ "beam_size": 4,
75
+ "lp_rate": 0.6,
76
+ "max_decoded_trg_len": 100
77
+ }
78
+
79
+ }