Hugging Face's logo

jcmc
/
wav2vec2-xls-r-1b-ir

Automatic Speech Recognition
Transformers
PyTorch
wav2vec2
mozilla-foundation/common_voice_7_0
Generated from Trainer
Model card Files
xet
 Community
jcmc commited on
Commit
3992870
·
1 Parent(s): 2b29e30

update model card README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -58
README.md CHANGED
@@ -20,8 +20,8 @@ should probably proofread and complete it, then remove this comment. -->
20
 
21
  This model is a fine-tuned version of [facebook/wav2vec2-xls-r-1b](https://huggingface.co/facebook/wav2vec2-xls-r-1b) on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - GA-IE dataset.
22
  It achieves the following results on the evaluation set:
23
- - Loss: nan
24
- - Wer: 1.0
25
 
26
  ## Model description
27
 
@@ -41,11 +41,11 @@ More information needed
41
 
42
  The following hyperparameters were used during training:
43
  - learning_rate: 5e-05
44
- - train_batch_size: 1
45
- - eval_batch_size: 1
46
  - seed: 42
47
  - gradient_accumulation_steps: 4
48
- - total_train_batch_size: 4
49
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
50
  - lr_scheduler_type: linear
51
  - lr_scheduler_warmup_steps: 500
@@ -54,59 +54,14 @@ The following hyperparameters were used during training:
54
 
55
  ### Training results
56
 
57
- | Training Loss | Epoch | Step | Validation Loss | Wer |
58
- |:-------------:|:-----:|:-----:|:---------------:|:---:|
59
- | 3.0395 | 1.94 | 500 | 3.0831 | 1.0 |
60
- | 3.0126 | 3.87 | 1000 | 2.9935 | 1.0 |
61
- | 2.9259 | 5.81 | 1500 | 2.9915 | 1.0 |
62
- | 2.9109 | 7.75 | 2000 | 2.9006 | 1.0 |
63
- | 2.8934 | 9.69 | 2500 | 2.9266 | 1.0 |
64
- | 2.9014 | 11.63 | 3000 | 2.8970 | 1.0 |
65
- | 2.8932 | 13.56 | 3500 | 2.8874 | 1.0 |
66
- | 0.0 | 15.5 | 4000 | nan | 1.0 |
67
- | 0.0 | 17.44 | 4500 | nan | 1.0 |
68
- | 0.0 | 19.38 | 5000 | nan | 1.0 |
69
- | 0.0 | 21.32 | 5500 | nan | 1.0 |
70
- | 0.0 | 23.26 | 6000 | nan | 1.0 |
71
- | 0.0 | 25.19 | 6500 | nan | 1.0 |
72
- | 0.0 | 27.13 | 7000 | nan | 1.0 |
73
- | 0.0 | 29.07 | 7500 | nan | 1.0 |
74
- | 0.0 | 31.01 | 8000 | nan | 1.0 |
75
- | 0.0 | 32.94 | 8500 | nan | 1.0 |
76
- | 0.0 | 34.88 | 9000 | nan | 1.0 |
77
- | 0.0 | 36.82 | 9500 | nan | 1.0 |
78
- | 0.0 | 38.76 | 10000 | nan | 1.0 |
79
- | 0.0 | 40.7 | 10500 | nan | 1.0 |
80
- | 0.0 | 42.63 | 11000 | nan | 1.0 |
81
- | 0.0 | 44.57 | 11500 | nan | 1.0 |
82
- | 0.0 | 46.51 | 12000 | nan | 1.0 |
83
- | 0.0 | 48.45 | 12500 | nan | 1.0 |
84
- | 0.0 | 50.39 | 13000 | nan | 1.0 |
85
- | 0.0 | 52.32 | 13500 | nan | 1.0 |
86
- | 0.0 | 54.26 | 14000 | nan | 1.0 |
87
- | 0.0 | 56.2 | 14500 | nan | 1.0 |
88
- | 0.0 | 58.14 | 15000 | nan | 1.0 |
89
- | 0.0 | 60.08 | 15500 | nan | 1.0 |
90
- | 0.0 | 62.02 | 16000 | nan | 1.0 |
91
- | 0.0 | 63.95 | 16500 | nan | 1.0 |
92
- | 0.0 | 65.89 | 17000 | nan | 1.0 |
93
- | 0.0 | 67.83 | 17500 | nan | 1.0 |
94
- | 0.0 | 69.77 | 18000 | nan | 1.0 |
95
- | 0.0 | 71.7 | 18500 | nan | 1.0 |
96
- | 0.0 | 73.64 | 19000 | nan | 1.0 |
97
- | 0.0 | 75.58 | 19500 | nan | 1.0 |
98
- | 0.0 | 77.52 | 20000 | nan | 1.0 |
99
- | 0.0 | 79.46 | 20500 | nan | 1.0 |
100
- | 0.0 | 81.39 | 21000 | nan | 1.0 |
101
- | 0.0 | 83.33 | 21500 | nan | 1.0 |
102
- | 0.0 | 85.27 | 22000 | nan | 1.0 |
103
- | 0.0 | 87.21 | 22500 | nan | 1.0 |
104
- | 0.0 | 89.15 | 23000 | nan | 1.0 |
105
- | 0.0 | 91.09 | 23500 | nan | 1.0 |
106
- | 0.0 | 93.02 | 24000 | nan | 1.0 |
107
- | 0.0 | 94.96 | 24500 | nan | 1.0 |
108
- | 0.0 | 96.9 | 25000 | nan | 1.0 |
109
- | 0.0 | 98.83 | 25500 | nan | 1.0 |
110
 
111
 
112
  ### Framework versions
 
20
 
21
  This model is a fine-tuned version of [facebook/wav2vec2-xls-r-1b](https://huggingface.co/facebook/wav2vec2-xls-r-1b) on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - GA-IE dataset.
22
  It achieves the following results on the evaluation set:
23
+ - Loss: 1.6569
24
+ - Wer: 0.8623
25
 
26
  ## Model description
27
 
 
41
 
42
  The following hyperparameters were used during training:
43
  - learning_rate: 5e-05
44
+ - train_batch_size: 8
45
+ - eval_batch_size: 8
46
  - seed: 42
47
  - gradient_accumulation_steps: 4
48
+ - total_train_batch_size: 32
49
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
50
  - lr_scheduler_type: linear
51
  - lr_scheduler_warmup_steps: 500
 
54
 
55
  ### Training results
56
 
57
+ | Training Loss | Epoch | Step | Validation Loss | Wer |
58
+ |:-------------:|:-----:|:----:|:---------------:|:------:|
59
+ | 2.1851 | 15.62 | 500 | 1.8067 | 0.9256 |
60
+ | 2.1586 | 31.25 | 1000 | 1.7883 | 0.9180 |
61
+ | 2.0302 | 46.86 | 1500 | 1.7571 | 0.9192 |
62
+ | 1.8706 | 62.49 | 2000 | 1.6314 | 0.8858 |
63
+ | 1.7008 | 78.12 | 2500 | 1.6131 | 0.8679 |
64
+ | 1.4982 | 93.74 | 3000 | 1.6540 | 0.8650 |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
65
 
66
 
67
  ### Framework versions