Hugging Face's logo

UrukHan
/
wav2vec2-russian

Automatic Speech Recognition
Transformers
PyTorch
TensorBoard
Safetensors
wav2vec2
Generated from Trainer
Inference Endpoints
Model card Files Metrics Community
4
Alikhan Urumov commited on
Commit
c934aaa
·
1 Parent(s): 20e08f8

update model card README.md

Browse files
Files changed (1) hide show
  1. README.md +28 -52
README.md CHANGED
@@ -14,8 +14,8 @@ should probably proofread and complete it, then remove this comment. -->
14
 
15
  This model is a fine-tuned version of [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on the None dataset.
16
  It achieves the following results on the evaluation set:
17
- - Loss: 1.2210
18
- - Wer: 0.4966
19
 
20
  ## Model description
21
 
@@ -34,68 +34,44 @@ More information needed
34
  ### Training hyperparameters
35
 
36
  The following hyperparameters were used during training:
37
- - learning_rate: 0.0001
38
  - train_batch_size: 16
39
  - eval_batch_size: 8
40
  - seed: 42
41
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
42
  - lr_scheduler_type: linear
43
  - lr_scheduler_warmup_steps: 1000
44
- - num_epochs: 12
45
  - mixed_precision_training: Native AMP
46
 
47
  ### Training results
48
 
49
  | Training Loss | Epoch | Step | Validation Loss | Wer |
50
  |:-------------:|:-----:|:-----:|:---------------:|:------:|
51
- | 5.0548 | 0.25 | 500 | 4.1857 | 0.9999 |
52
- | 3.0204 | 0.5 | 1000 | 1.9996 | 0.9998 |
53
- | 1.8692 | 0.74 | 1500 | 1.6426 | 0.8698 |
54
- | 1.5154 | 0.99 | 2000 | 1.6156 | 0.7481 |
55
- | 1.3677 | 1.24 | 2500 | 2.1281 | 0.7120 |
56
- | 1.3223 | 1.49 | 3000 | 1.5192 | 0.6846 |
57
- | 1.2512 | 1.73 | 3500 | 1.0993 | 0.6634 |
58
- | 1.2257 | 1.98 | 4000 | 1.1039 | 0.6493 |
59
- | 1.1418 | 2.23 | 4500 | 1.0170 | 0.6241 |
60
- | 1.1213 | 2.48 | 5000 | 0.8436 | 0.6191 |
61
- | 1.112 | 2.73 | 5500 | 0.7326 | 0.6102 |
62
- | 1.0912 | 2.97 | 6000 | 0.7054 | 0.5976 |
63
- | 1.0465 | 3.22 | 6500 | 1.0887 | 0.5941 |
64
- | 1.0215 | 3.47 | 7000 | 1.4577 | 0.5793 |
65
- | 1.0244 | 3.72 | 7500 | 1.6058 | 0.5855 |
66
- | 1.0254 | 3.96 | 8000 | 1.3366 | 0.5750 |
67
- | 0.9558 | 4.21 | 8500 | 0.8088 | 0.5644 |
68
- | 0.966 | 4.46 | 9000 | 0.9650 | 0.5636 |
69
- | 0.9674 | 4.71 | 9500 | 0.9047 | 0.5532 |
70
- | 0.9373 | 4.96 | 10000 | 1.0342 | 0.5422 |
71
- | 0.9126 | 5.2 | 10500 | 1.2346 | 0.5462 |
72
- | 0.9063 | 5.45 | 11000 | 1.2696 | 0.5412 |
73
- | 0.9126 | 5.7 | 11500 | 1.4693 | 0.5317 |
74
- | 0.8936 | 5.95 | 12000 | 1.9096 | 0.5369 |
75
- | 0.8621 | 6.19 | 12500 | 1.6382 | 0.5326 |
76
- | 0.8695 | 6.44 | 13000 | 0.9466 | 0.5252 |
77
- | 0.8423 | 6.69 | 13500 | 1.6286 | 0.5355 |
78
- | 0.8494 | 6.94 | 14000 | 0.8368 | 0.5264 |
79
- | 0.8354 | 7.19 | 14500 | 0.6893 | 0.5216 |
80
- | 0.8133 | 7.43 | 15000 | 0.5916 | 0.5175 |
81
- | 0.8147 | 7.68 | 15500 | 0.7813 | 0.5221 |
82
- | 0.8258 | 7.93 | 16000 | 1.3814 | 0.5129 |
83
- | 0.8079 | 8.18 | 16500 | 0.8368 | 0.5176 |
84
- | 0.7868 | 8.42 | 17000 | 0.9456 | 0.5159 |
85
- | 0.7955 | 8.67 | 17500 | 0.7412 | 0.5170 |
86
- | 0.7921 | 8.92 | 18000 | 0.6256 | 0.5066 |
87
- | 0.7536 | 9.17 | 18500 | 0.8792 | 0.5101 |
88
- | 0.7667 | 9.42 | 19000 | 1.0615 | 0.5032 |
89
- | 0.772 | 9.66 | 19500 | 1.1312 | 0.5086 |
90
- | 0.7418 | 9.91 | 20000 | 1.3485 | 0.4990 |
91
- | 0.7577 | 10.16 | 20500 | 1.0788 | 0.5037 |
92
- | 0.7311 | 10.41 | 21000 | 0.9978 | 0.5032 |
93
- | 0.7419 | 10.65 | 21500 | 1.3925 | 0.5017 |
94
- | 0.74 | 10.9 | 22000 | 1.4191 | 0.4981 |
95
- | 0.7297 | 11.15 | 22500 | 1.1082 | 0.4994 |
96
- | 0.737 | 11.4 | 23000 | 1.1208 | 0.4971 |
97
- | 0.7266 | 11.65 | 23500 | 1.1595 | 0.4952 |
98
- | 0.7091 | 11.89 | 24000 | 1.2210 | 0.4966 |
99
 
100
 
101
  ### Framework versions
 
14
 
15
  This model is a fine-tuned version of [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on the None dataset.
16
  It achieves the following results on the evaluation set:
17
+ - Loss: 1.2972
18
+ - Wer: 0.4920
19
 
20
  ## Model description
21
 
 
34
  ### Training hyperparameters
35
 
36
  The following hyperparameters were used during training:
37
+ - learning_rate: 1e-05
38
  - train_batch_size: 16
39
  - eval_batch_size: 8
40
  - seed: 42
41
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
42
  - lr_scheduler_type: linear
43
  - lr_scheduler_warmup_steps: 1000
44
+ - num_epochs: 6
45
  - mixed_precision_training: Native AMP
46
 
47
  ### Training results
48
 
49
  | Training Loss | Epoch | Step | Validation Loss | Wer |
50
  |:-------------:|:-----:|:-----:|:---------------:|:------:|
51
+ | 0.6885 | 0.25 | 500 | 1.2398 | 0.4926 |
52
+ | 0.6499 | 0.5 | 1000 | 1.2338 | 0.4957 |
53
+ | 0.6446 | 0.74 | 1500 | 1.2716 | 0.4975 |
54
+ | 0.6966 | 0.99 | 2000 | 1.5439 | 0.4942 |
55
+ | 0.7188 | 1.24 | 2500 | 1.4889 | 0.4960 |
56
+ | 0.7116 | 1.49 | 3000 | 1.0075 | 0.4921 |
57
+ | 0.7178 | 1.73 | 3500 | 1.3123 | 0.4984 |
58
+ | 0.7034 | 1.98 | 4000 | 1.1037 | 0.4919 |
59
+ | 0.7116 | 2.23 | 4500 | 0.9947 | 0.4942 |
60
+ | 0.7203 | 2.48 | 5000 | 1.1547 | 0.4940 |
61
+ | 0.72 | 2.73 | 5500 | 1.1245 | 0.4960 |
62
+ | 0.6921 | 2.97 | 6000 | 1.0844 | 0.4947 |
63
+ | 0.702 | 3.22 | 6500 | 1.1999 | 0.4931 |
64
+ | 0.6965 | 3.47 | 7000 | 1.2106 | 0.4906 |
65
+ | 0.6928 | 3.72 | 7500 | 1.1275 | 0.4939 |
66
+ | 0.6973 | 3.96 | 8000 | 1.1953 | 0.4922 |
67
+ | 0.7065 | 4.21 | 8500 | 1.2046 | 0.4931 |
68
+ | 0.6921 | 4.46 | 9000 | 1.2052 | 0.4926 |
69
+ | 0.6977 | 4.71 | 9500 | 1.2806 | 0.4922 |
70
+ | 0.688 | 4.96 | 10000 | 1.3327 | 0.4926 |
71
+ | 0.7003 | 5.2 | 10500 | 1.3518 | 0.4930 |
72
+ | 0.6981 | 5.45 | 11000 | 1.3295 | 0.4936 |
73
+ | 0.6913 | 5.7 | 11500 | 1.3065 | 0.4928 |
74
+ | 0.6822 | 5.95 | 12000 | 1.2972 | 0.4920 |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
75
 
76
 
77
  ### Framework versions