Hugging Face's logo

AhBotNLP
/
wav2vec2-xls-r-164m-id

Automatic Speech Recognition
Transformers
PyTorch
wav2vec2
Generated from Trainer
Eval Results (legacy)
Model card Files
xet
 Community
1
evanarlian commited on
Commit
facdc9c
·
1 Parent(s): 2258d47

update model card README.md

Browse files
Files changed (1) hide show
  1. README.md +24 -34
README.md CHANGED
@@ -17,7 +17,7 @@ model-index:
17
  metrics:
18
  - name: Wer
19
  type: wer
20
- value: 0.2990499031454663
21
  ---
22
 
23
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
@@ -25,10 +25,10 @@ should probably proofread and complete it, then remove this comment. -->
25
 
26
  # wav2vec2-xls-r-164m-id
27
 
28
- This model is a fine-tuned version of [evanarlian/wav2vec2-xls-r-164m-id](https://huggingface.co/evanarlian/wav2vec2-xls-r-164m-id) on the evanarlian/common_voice_11_0_id_filtered dataset.
29
  It achieves the following results on the evaluation set:
30
- - Loss: 0.3510
31
- - Wer: 0.2990
32
 
33
  ## Model description
34
 
@@ -47,47 +47,37 @@ More information needed
47
  ### Training hyperparameters
48
 
49
  The following hyperparameters were used during training:
50
- - learning_rate: 5e-05
51
  - train_batch_size: 24
52
  - eval_batch_size: 24
53
  - seed: 42
54
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
55
  - lr_scheduler_type: linear
56
- - lr_scheduler_warmup_ratio: 0.2
57
- - num_epochs: 50.0
58
  - mixed_precision_training: Native AMP
59
 
60
  ### Training results
61
 
62
  | Training Loss | Epoch | Step | Validation Loss | Wer |
63
  |:-------------:|:-----:|:-----:|:---------------:|:------:|
64
- | 0.089 | 1.84 | 2000 | 0.3205 | 0.3168 |
65
- | 0.0882 | 3.67 | 4000 | 0.3243 | 0.3203 |
66
- | 0.0868 | 5.51 | 6000 | 0.3272 | 0.3183 |
67
- | 0.0926 | 7.35 | 8000 | 0.3365 | 0.3209 |
68
- | 0.0943 | 9.18 | 10000 | 0.3400 | 0.3221 |
69
- | 0.0979 | 11.02 | 12000 | 0.3269 | 0.3192 |
70
- | 0.09 | 12.86 | 14000 | 0.3384 | 0.3164 |
71
- | 0.0877 | 14.69 | 16000 | 0.3284 | 0.3183 |
72
- | 0.0808 | 16.53 | 18000 | 0.3366 | 0.3189 |
73
- | 0.0835 | 18.37 | 20000 | 0.3306 | 0.3156 |
74
- | 0.08 | 20.2 | 22000 | 0.3384 | 0.3133 |
75
- | 0.0806 | 22.04 | 24000 | 0.3307 | 0.3109 |
76
- | 0.0749 | 23.88 | 26000 | 0.3493 | 0.3118 |
77
- | 0.073 | 25.71 | 28000 | 0.3479 | 0.3088 |
78
- | 0.0754 | 27.55 | 30000 | 0.3482 | 0.3109 |
79
- | 0.0697 | 29.38 | 32000 | 0.3515 | 0.3090 |
80
- | 0.07 | 31.22 | 34000 | 0.3532 | 0.3101 |
81
- | 0.0672 | 33.06 | 36000 | 0.3668 | 0.3086 |
82
- | 0.0713 | 34.89 | 38000 | 0.3560 | 0.3048 |
83
- | 0.0637 | 36.73 | 40000 | 0.3522 | 0.3028 |
84
- | 0.0695 | 38.57 | 42000 | 0.3407 | 0.3014 |
85
- | 0.0657 | 40.4 | 44000 | 0.3456 | 0.3025 |
86
- | 0.0598 | 42.24 | 46000 | 0.3498 | 0.3013 |
87
- | 0.059 | 44.08 | 48000 | 0.3563 | 0.3012 |
88
- | 0.0645 | 45.91 | 50000 | 0.3514 | 0.3002 |
89
- | 0.0595 | 47.75 | 52000 | 0.3545 | 0.3000 |
90
- | 0.064 | 49.59 | 54000 | 0.3510 | 0.2990 |
91
 
92
 
93
  ### Framework versions
 
17
  metrics:
18
  - name: Wer
19
  type: wer
20
+ value: 0.2923162069919749
21
  ---
22
 
23
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
 
25
 
26
  # wav2vec2-xls-r-164m-id
27
 
28
+ This model is a fine-tuned version of [evanarlian/distil-wav2vec2-xls-r-164m-id](https://huggingface.co/evanarlian/distil-wav2vec2-xls-r-164m-id) on the evanarlian/common_voice_11_0_id_filtered dataset.
29
  It achieves the following results on the evaluation set:
30
+ - Loss: 0.2865
31
+ - Wer: 0.2923
32
 
33
  ## Model description
34
 
 
47
  ### Training hyperparameters
48
 
49
  The following hyperparameters were used during training:
50
+ - learning_rate: 0.0001
51
  - train_batch_size: 24
52
  - eval_batch_size: 24
53
  - seed: 42
54
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
55
  - lr_scheduler_type: linear
56
+ - lr_scheduler_warmup_ratio: 0.3
57
+ - num_epochs: 80.0
58
  - mixed_precision_training: Native AMP
59
 
60
  ### Training results
61
 
62
  | Training Loss | Epoch | Step | Validation Loss | Wer |
63
  |:-------------:|:-----:|:-----:|:---------------:|:------:|
64
+ | 1.4047 | 4.59 | 5000 | 1.0167 | 0.9138 |
65
+ | 0.587 | 9.18 | 10000 | 0.4639 | 0.5615 |
66
+ | 0.3782 | 13.77 | 15000 | 0.3375 | 0.4496 |
67
+ | 0.2867 | 18.37 | 20000 | 0.2881 | 0.4022 |
68
+ | 0.2519 | 22.96 | 25000 | 0.2775 | 0.3700 |
69
+ | 0.1941 | 27.55 | 30000 | 0.2701 | 0.3516 |
70
+ | 0.1727 | 32.14 | 35000 | 0.2795 | 0.3486 |
71
+ | 0.1448 | 36.73 | 40000 | 0.2878 | 0.3364 |
72
+ | 0.1251 | 41.32 | 45000 | 0.2649 | 0.3275 |
73
+ | 0.113 | 45.91 | 50000 | 0.2862 | 0.3168 |
74
+ | 0.0994 | 50.51 | 55000 | 0.2798 | 0.3091 |
75
+ | 0.0938 | 55.1 | 60000 | 0.2864 | 0.3070 |
76
+ | 0.0853 | 59.69 | 65000 | 0.2860 | 0.3069 |
77
+ | 0.0724 | 64.28 | 70000 | 0.2994 | 0.3003 |
78
+ | 0.0723 | 68.87 | 75000 | 0.2951 | 0.2983 |
79
+ | 0.0666 | 73.46 | 80000 | 0.2886 | 0.2941 |
80
+ | 0.0659 | 78.05 | 85000 | 0.2865 | 0.2923 |
 
 
 
 
 
 
 
 
 
 
81
 
82
 
83
  ### Framework versions