Hugging Face's logo

alexue4
/
text-normalization-ru-new

Text Generation
Transformers
PyTorch
TensorBoard
Safetensors
Russian
t5
text2text-generation
Generated from Trainer
text-generation-inference
Model card Files
xet
Metrics Community
alexue4 commited on
Commit
25a79b1
·
1 Parent(s): 407d7de

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -105
README.md CHANGED
@@ -1,19 +1,18 @@
1
  ---
2
  license: mit
3
- base_model: cointegrated/rut5-small
 
 
4
  tags:
5
- - generated_from_trainer
6
- model-index:
7
- - name: text-normalization-ru-new
8
- results: []
9
  ---
10
 
11
- <!-- This model card has been generated automatically according to the information the Trainer had access to. You
12
- should probably proofread and complete it, then remove this comment. -->
13
-
14
  # text-normalization-ru-new
 
 
 
 
15
 
16
- This model is a fine-tuned version of [cointegrated/rut5-small](https://huggingface.co/cointegrated/rut5-small) on an unknown dataset.
17
  It achieves the following results on the evaluation set:
18
  - Loss: 0.0177
19
  - Mean Distance: 0
@@ -21,99 +20,8 @@ It achieves the following results on the evaluation set:
21
 
22
  ## Model description
23
 
24
- More information needed
25
-
26
- ## Intended uses & limitations
27
-
28
- More information needed
29
-
30
- ## Training and evaluation data
31
-
32
- More information needed
33
-
34
- ## Training procedure
35
-
36
- ### Training hyperparameters
37
-
38
- The following hyperparameters were used during training:
39
- - learning_rate: 0.001
40
- - train_batch_size: 30
41
- - eval_batch_size: 30
42
- - seed: 42
43
- - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
44
- - lr_scheduler_type: linear
45
- - lr_scheduler_warmup_ratio: 0.1
46
- - num_epochs: 60
47
-
48
- ### Training results
49
-
50
- | Training Loss | Epoch | Step | Validation Loss | Mean Distance | Max Distance |
51
- |:-------------:|:-----:|:------:|:---------------:|:-------------:|:------------:|
52
- | 0.2236 | 1.0 | 3298 | 0.1120 | 5 | 133 |
53
- | 0.1179 | 2.0 | 6596 | 0.0548 | 3 | 82 |
54
- | 0.0829 | 3.0 | 9894 | 0.0425 | 1 | 46 |
55
- | 0.0643 | 4.0 | 13192 | 0.0311 | 1 | 64 |
56
- | 0.0538 | 5.0 | 16490 | 0.0267 | 1 | 48 |
57
- | 0.0469 | 6.0 | 19788 | 0.0396 | 2 | 80 |
58
- | 0.0385 | 7.0 | 23086 | 0.0262 | 2 | 73 |
59
- | 0.0316 | 8.0 | 26384 | 0.0223 | 1 | 40 |
60
- | 0.0263 | 9.0 | 29682 | 0.0240 | 1 | 69 |
61
- | 0.0226 | 10.0 | 32980 | 0.0203 | 1 | 60 |
62
- | 0.0203 | 11.0 | 36278 | 0.0177 | 1 | 54 |
63
- | 0.0178 | 12.0 | 39576 | 0.0188 | 1 | 61 |
64
- | 0.0154 | 13.0 | 42874 | 0.0296 | 1 | 65 |
65
- | 0.0138 | 14.0 | 46172 | 0.0201 | 1 | 55 |
66
- | 0.012 | 15.0 | 49470 | 0.0268 | 1 | 67 |
67
- | 0.0109 | 16.0 | 52768 | 0.0163 | 1 | 35 |
68
- | 0.0105 | 17.0 | 56066 | 0.0136 | 1 | 26 |
69
- | 0.0092 | 18.0 | 59364 | 0.0202 | 1 | 65 |
70
- | 0.0087 | 19.0 | 62662 | 0.0221 | 1 | 65 |
71
- | 0.0075 | 20.0 | 65960 | 0.0203 | 1 | 33 |
72
- | 0.0067 | 21.0 | 69258 | 0.0226 | 1 | 26 |
73
- | 0.0062 | 22.0 | 72556 | 0.0184 | 1 | 24 |
74
- | 0.0059 | 23.0 | 75854 | 0.0131 | 0 | 18 |
75
- | 0.0054 | 24.0 | 79152 | 0.0270 | 1 | 58 |
76
- | 0.0052 | 25.0 | 82450 | 0.0244 | 1 | 45 |
77
- | 0.0044 | 26.0 | 85748 | 0.0149 | 1 | 23 |
78
- | 0.0043 | 27.0 | 89046 | 0.0256 | 1 | 63 |
79
- | 0.0038 | 28.0 | 92344 | 0.0172 | 1 | 30 |
80
- | 0.0036 | 29.0 | 95642 | 0.0224 | 1 | 37 |
81
- | 0.0033 | 30.0 | 98940 | 0.0194 | 1 | 30 |
82
- | 0.0031 | 31.0 | 102238 | 0.0238 | 1 | 59 |
83
- | 0.003 | 32.0 | 105536 | 0.0200 | 1 | 28 |
84
- | 0.0028 | 33.0 | 108834 | 0.0161 | 0 | 18 |
85
- | 0.0027 | 34.0 | 112132 | 0.0215 | 1 | 26 |
86
- | 0.0025 | 35.0 | 115430 | 0.0198 | 0 | 19 |
87
- | 0.0023 | 36.0 | 118728 | 0.0168 | 0 | 24 |
88
- | 0.002 | 37.0 | 122026 | 0.0221 | 1 | 32 |
89
- | 0.0019 | 38.0 | 125324 | 0.0214 | 1 | 32 |
90
- | 0.0017 | 39.0 | 128622 | 0.0186 | 0 | 19 |
91
- | 0.0017 | 40.0 | 131920 | 0.0171 | 0 | 23 |
92
- | 0.0016 | 41.0 | 135218 | 0.0164 | 0 | 17 |
93
- | 0.0015 | 42.0 | 138516 | 0.0166 | 1 | 21 |
94
- | 0.0014 | 43.0 | 141814 | 0.0167 | 0 | 21 |
95
- | 0.0019 | 44.0 | 145112 | 0.0192 | 1 | 32 |
96
- | 0.0011 | 45.0 | 148410 | 0.0209 | 1 | 27 |
97
- | 0.0011 | 46.0 | 151708 | 0.0218 | 0 | 23 |
98
- | 0.001 | 47.0 | 155006 | 0.0195 | 0 | 25 |
99
- | 0.0009 | 48.0 | 158304 | 0.0166 | 0 | 15 |
100
- | 0.0008 | 49.0 | 161602 | 0.0210 | 1 | 31 |
101
- | 0.0008 | 50.0 | 164900 | 0.0230 | 0 | 22 |
102
- | 0.0008 | 51.0 | 168198 | 0.0184 | 0 | 15 |
103
- | 0.0007 | 52.0 | 171496 | 0.0183 | 0 | 15 |
104
- | 0.0006 | 53.0 | 174794 | 0.0234 | 1 | 32 |
105
- | 0.0005 | 54.0 | 178092 | 0.0227 | 0 | 24 |
106
- | 0.0004 | 55.0 | 181390 | 0.0188 | 0 | 15 |
107
- | 0.0005 | 56.0 | 184688 | 0.0191 | 0 | 15 |
108
- | 0.0004 | 57.0 | 187986 | 0.0183 | 0 | 15 |
109
- | 0.0003 | 58.0 | 191284 | 0.0180 | 0 | 15 |
110
- | 0.0003 | 59.0 | 194582 | 0.0180 | 0 | 15 |
111
- | 0.0004 | 60.0 | 197880 | 0.0177 | 0 | 15 |
112
-
113
-
114
- ### Framework versions
115
-
116
- - Transformers 4.32.1
117
- - Pytorch 2.0.1+cu117
118
- - Datasets 2.14.4
119
- - Tokenizers 0.13.3
 
1
  ---
2
  license: mit
3
+ language:
4
+ - ru
5
+ library_name: transformers
6
  tags:
7
+ - text-generation-inference
 
 
 
8
  ---
9
 
 
 
 
10
  # text-normalization-ru-new
11
+ Normalization for Russian text. Couldn't find any existing solutions (besides algorithms, don't like those) so made this.
12
+ It was designed for Silero TTS model which cant handle english and numbers for russian text to speach.
13
+
14
+ This model is a fine-tuned version of [cointegrated/rut5-small](https://huggingface.co/cointegrated/rut5-small) on https://www.kaggle.com/c/text-normalization-challenge-russian-language and additional dataset prepared by me using typical messages.
15
 
 
16
  It achieves the following results on the evaluation set:
17
  - Loss: 0.0177
18
  - Mean Distance: 0
 
20
 
21
  ## Model description
22
 
23
+ Tiny T5 trained from scratch for normalizing Russian texts:
24
+ - translating numbers into words
25
+ - expanding abbreviations into phonetic letter combinations
26
+ - transliterating english into russian letters
27
+ - whatever else was in the dataset (see below)