patrickvonplaten
commited on
Commit
•
c613557
1
Parent(s):
97157ad
update model card README.md
Browse files
README.md
CHANGED
@@ -20,9 +20,9 @@ should probably proofread and complete it, then remove this comment. -->
|
|
20 |
|
21 |
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the COMMON_VOICE - TR dataset.
|
22 |
It achieves the following results on the evaluation set:
|
23 |
-
- Loss: 0.
|
24 |
-
- Wer: 0.
|
25 |
-
- Cer: 0.
|
26 |
|
27 |
## Model description
|
28 |
|
@@ -41,45 +41,29 @@ More information needed
|
|
41 |
### Training hyperparameters
|
42 |
|
43 |
The following hyperparameters were used during training:
|
44 |
-
- learning_rate: 0.
|
45 |
-
- train_batch_size:
|
46 |
- eval_batch_size: 8
|
47 |
- seed: 42
|
48 |
- distributed_type: multi-GPU
|
49 |
- num_devices: 8
|
50 |
-
- total_train_batch_size:
|
51 |
- total_eval_batch_size: 64
|
52 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
53 |
- lr_scheduler_type: linear
|
54 |
- lr_scheduler_warmup_steps: 500
|
55 |
-
- num_epochs:
|
56 |
- mixed_precision_training: Native AMP
|
57 |
|
58 |
### Training results
|
59 |
|
60 |
-
| Training Loss | Epoch | Step
|
61 |
-
|
62 |
-
|
|
63 |
-
|
|
64 |
-
|
|
65 |
-
|
|
66 |
-
|
|
67 |
-
| 0.8303 | 27.52 | 3000 | 0.3826 | 0.5113 | 0.1292 |
|
68 |
-
| 0.9115 | 32.11 | 3500 | 0.3819 | 0.4324 | 0.1070 |
|
69 |
-
| 0.8193 | 36.7 | 4000 | 0.3694 | 0.4223 | 0.1036 |
|
70 |
-
| 0.8948 | 41.28 | 4500 | 0.3714 | 0.4100 | 0.1005 |
|
71 |
-
| 0.774 | 45.87 | 5000 | 0.3558 | 0.3923 | 0.0971 |
|
72 |
-
| 0.8194 | 50.46 | 5500 | 0.3729 | 0.4603 | 0.1180 |
|
73 |
-
| 0.8616 | 55.05 | 6000 | 0.3616 | 0.3908 | 0.0963 |
|
74 |
-
| 0.7901 | 59.63 | 6500 | 0.3575 | 0.3837 | 0.0952 |
|
75 |
-
| 0.778 | 64.22 | 7000 | 0.3732 | 0.3790 | 0.0928 |
|
76 |
-
| 0.7238 | 68.81 | 7500 | 0.3674 | 0.3734 | 0.0904 |
|
77 |
-
| 0.6985 | 73.39 | 8000 | 0.3627 | 0.3615 | 0.0863 |
|
78 |
-
| 0.5889 | 77.98 | 8500 | 0.3705 | 0.3548 | 0.0858 |
|
79 |
-
| 0.5447 | 82.57 | 9000 | 0.3678 | 0.3534 | 0.0854 |
|
80 |
-
| 0.4763 | 87.16 | 9500 | 0.3627 | 0.3509 | 0.0840 |
|
81 |
-
| 0.3544 | 91.74 | 10000 | 0.3690 | 0.3495 | 0.0834 |
|
82 |
-
| 0.4879 | 96.33 | 10500 | 0.3683 | 0.3418 | 0.0820 |
|
83 |
|
84 |
|
85 |
### Framework versions
|
|
|
20 |
|
21 |
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the COMMON_VOICE - TR dataset.
|
22 |
It achieves the following results on the evaluation set:
|
23 |
+
- Loss: 0.3527
|
24 |
+
- Wer: 0.3212
|
25 |
+
- Cer: 0.0750
|
26 |
|
27 |
## Model description
|
28 |
|
|
|
41 |
### Training hyperparameters
|
42 |
|
43 |
The following hyperparameters were used during training:
|
44 |
+
- learning_rate: 0.0005
|
45 |
+
- train_batch_size: 8
|
46 |
- eval_batch_size: 8
|
47 |
- seed: 42
|
48 |
- distributed_type: multi-GPU
|
49 |
- num_devices: 8
|
50 |
+
- total_train_batch_size: 64
|
51 |
- total_eval_batch_size: 64
|
52 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
53 |
- lr_scheduler_type: linear
|
54 |
- lr_scheduler_warmup_steps: 500
|
55 |
+
- num_epochs: 50.0
|
56 |
- mixed_precision_training: Native AMP
|
57 |
|
58 |
### Training results
|
59 |
|
60 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|
61 |
+
|:-------------:|:-----:|:----:|:---------------:|:------:|:------:|
|
62 |
+
| 0.7897 | 9.09 | 500 | 0.4873 | 0.6006 | 0.1491 |
|
63 |
+
| 0.5166 | 18.18 | 1000 | 0.4058 | 0.4238 | 0.1070 |
|
64 |
+
| 0.4012 | 27.27 | 1500 | 0.3716 | 0.3781 | 0.0935 |
|
65 |
+
| 0.3851 | 36.36 | 2000 | 0.3565 | 0.3458 | 0.0836 |
|
66 |
+
| 0.2212 | 45.45 | 2500 | 0.3580 | 0.3272 | 0.0766 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
|
68 |
|
69 |
### Framework versions
|