Hugging Face's logo

adrianSauer
/
wav2vec2-wer-extension

Automatic Speech Recognition
Transformers
TensorBoard
Safetensors
Guaraní
wav2vec2
Generated from Trainer
Eval Results
Inference Endpoints
Model card Files Metrics Community
adrianSauer commited on
Commit
ffee101
·
verified ·
1 Parent(s): 80a0028

End of training

Browse files
Files changed (1) hide show
  1. README.md +14 -12
README.md CHANGED
@@ -1,4 +1,5 @@
1
  ---
 
2
  language:
3
  - gn
4
  license: apache-2.0
@@ -24,7 +25,7 @@ model-index:
24
  metrics:
25
  - name: Wer
26
  type: wer
27
- value: 45.839210155148095
28
  ---
29
 
30
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
@@ -34,8 +35,8 @@ should probably proofread and complete it, then remove this comment. -->
34
 
35
  This model is a fine-tuned version of [glob-asr/wav2vec2-large-xls-r-300m-guarani-small](https://huggingface.co/glob-asr/wav2vec2-large-xls-r-300m-guarani-small) on the Common Voice 16 dataset.
36
  It achieves the following results on the evaluation set:
37
- - Loss: 0.3327
38
- - Wer: 45.8392
39
 
40
  ## Model description
41
 
@@ -54,7 +55,7 @@ More information needed
54
  ### Training hyperparameters
55
 
56
  The following hyperparameters were used during training:
57
- - learning_rate: 1e-05
58
  - train_batch_size: 8
59
  - eval_batch_size: 16
60
  - seed: 42
@@ -62,24 +63,25 @@ The following hyperparameters were used during training:
62
  - total_train_batch_size: 16
63
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
64
  - lr_scheduler_type: constant_with_warmup
65
- - lr_scheduler_warmup_steps: 50
66
- - training_steps: 500
67
  - mixed_precision_training: Native AMP
68
 
69
  ### Training results
70
 
71
  | Training Loss | Epoch | Step | Validation Loss | Wer |
72
  |:-------------:|:------:|:----:|:---------------:|:-------:|
73
- | 1.397 | 0.0991 | 100 | 0.3675 | 48.4062 |
74
- | 1.0613 | 0.1982 | 200 | 0.3604 | 50.5219 |
75
- | 1.0365 | 0.2973 | 300 | 0.3500 | 48.8575 |
76
- | 0.9822 | 0.3964 | 400 | 0.3454 | 47.5599 |
77
- | 0.9197 | 0.4955 | 500 | 0.3327 | 45.8392 |
 
78
 
79
 
80
  ### Framework versions
81
 
82
- - Transformers 4.44.0
83
  - Pytorch 2.3.1+cu121
84
  - Datasets 2.21.0
85
  - Tokenizers 0.19.1
 
1
  ---
2
+ library_name: transformers
3
  language:
4
  - gn
5
  license: apache-2.0
 
25
  metrics:
26
  - name: Wer
27
  type: wer
28
+ value: 39.84010659560293
29
  ---
30
 
31
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
 
35
 
36
  This model is a fine-tuned version of [glob-asr/wav2vec2-large-xls-r-300m-guarani-small](https://huggingface.co/glob-asr/wav2vec2-large-xls-r-300m-guarani-small) on the Common Voice 16 dataset.
37
  It achieves the following results on the evaluation set:
38
+ - Loss: 0.2438
39
+ - Wer: 39.8401
40
 
41
  ## Model description
42
 
 
55
  ### Training hyperparameters
56
 
57
  The following hyperparameters were used during training:
58
+ - learning_rate: 5e-05
59
  - train_batch_size: 8
60
  - eval_batch_size: 16
61
  - seed: 42
 
63
  - total_train_batch_size: 16
64
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
65
  - lr_scheduler_type: constant_with_warmup
66
+ - lr_scheduler_warmup_steps: 3000
67
+ - training_steps: 3000
68
  - mixed_precision_training: Native AMP
69
 
70
  ### Training results
71
 
72
  | Training Loss | Epoch | Step | Validation Loss | Wer |
73
  |:-------------:|:------:|:----:|:---------------:|:-------:|
74
+ | 1.2579 | 0.4955 | 500 | 0.3710 | 53.4310 |
75
+ | 0.919 | 0.9911 | 1000 | 0.3295 | 49.9001 |
76
+ | 0.746 | 1.4866 | 1500 | 0.2902 | 45.1033 |
77
+ | 0.6767 | 1.9822 | 2000 | 0.2674 | 43.3711 |
78
+ | 0.574 | 2.4777 | 2500 | 0.2677 | 42.5716 |
79
+ | 0.5485 | 2.9732 | 3000 | 0.2438 | 39.8401 |
80
 
81
 
82
  ### Framework versions
83
 
84
+ - Transformers 4.44.1
85
  - Pytorch 2.3.1+cu121
86
  - Datasets 2.21.0
87
  - Tokenizers 0.19.1