patrickvonplaten
commited on
Commit
•
1c33bc7
1
Parent(s):
a51adf5
update model card README.md
Browse files
README.md
CHANGED
@@ -18,8 +18,8 @@ should probably proofread and complete it, then remove this comment. -->
|
|
18 |
|
19 |
This model is a fine-tuned version of [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on the MULTILINGUAL_LIBRISPEECH - GERMAN dataset.
|
20 |
It achieves the following results on the evaluation set:
|
21 |
-
- Loss: 0.
|
22 |
-
- Wer: 0.
|
23 |
|
24 |
## Model description
|
25 |
|
@@ -38,22 +38,33 @@ More information needed
|
|
38 |
### Training hyperparameters
|
39 |
|
40 |
The following hyperparameters were used during training:
|
41 |
-
- learning_rate: 0.
|
42 |
- train_batch_size: 32
|
43 |
- eval_batch_size: 8
|
44 |
- seed: 42
|
45 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
46 |
- lr_scheduler_type: linear
|
47 |
-
- lr_scheduler_warmup_steps:
|
48 |
-
- num_epochs:
|
49 |
- mixed_precision_training: Native AMP
|
50 |
|
51 |
### Training results
|
52 |
|
53 |
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
54 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
|
55 |
-
|
|
56 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
|
59 |
### Framework versions
|
|
|
18 |
|
19 |
This model is a fine-tuned version of [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on the MULTILINGUAL_LIBRISPEECH - GERMAN dataset.
|
20 |
It achieves the following results on the evaluation set:
|
21 |
+
- Loss: 0.1936
|
22 |
+
- Wer: 0.1316
|
23 |
|
24 |
## Model description
|
25 |
|
|
|
38 |
### Training hyperparameters
|
39 |
|
40 |
The following hyperparameters were used during training:
|
41 |
+
- learning_rate: 0.0001
|
42 |
- train_batch_size: 32
|
43 |
- eval_batch_size: 8
|
44 |
- seed: 42
|
45 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
46 |
- lr_scheduler_type: linear
|
47 |
+
- lr_scheduler_warmup_steps: 1000
|
48 |
+
- num_epochs: 100.0
|
49 |
- mixed_precision_training: Native AMP
|
50 |
|
51 |
### Training results
|
52 |
|
53 |
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
54 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
|
55 |
+
| 2.9888 | 7.25 | 500 | 2.9192 | 1.0 |
|
56 |
+
| 2.9313 | 14.49 | 1000 | 2.8698 | 1.0 |
|
57 |
+
| 1.068 | 21.74 | 1500 | 0.2647 | 0.2565 |
|
58 |
+
| 0.8151 | 28.99 | 2000 | 0.2067 | 0.1719 |
|
59 |
+
| 0.764 | 36.23 | 2500 | 0.1975 | 0.1568 |
|
60 |
+
| 0.7332 | 43.48 | 3000 | 0.1812 | 0.1463 |
|
61 |
+
| 0.5952 | 50.72 | 3500 | 0.1923 | 0.1428 |
|
62 |
+
| 0.6655 | 57.97 | 4000 | 0.1900 | 0.1404 |
|
63 |
+
| 0.574 | 65.22 | 4500 | 0.1822 | 0.1370 |
|
64 |
+
| 0.6211 | 72.46 | 5000 | 0.1937 | 0.1355 |
|
65 |
+
| 0.5883 | 79.71 | 5500 | 0.1872 | 0.1335 |
|
66 |
+
| 0.5666 | 86.96 | 6000 | 0.1874 | 0.1324 |
|
67 |
+
| 0.5436 | 94.2 | 6500 | 0.1924 | 0.1311 |
|
68 |
|
69 |
|
70 |
### Framework versions
|