Update README.md
Browse files
README.md
CHANGED
|
@@ -24,41 +24,34 @@ model-index:
|
|
| 24 |
value: 0.2884547694473777
|
| 25 |
---
|
| 26 |
|
| 27 |
-
|
| 28 |
-
This model is a fine-tuned version of Wav2Vec2 on the openslr dataset.
|
|
|
|
| 29 |
|
| 30 |
-
Loss: 0.2982
|
| 31 |
-
Wer: 0.2885
|
| 32 |
-
Model description
|
| 33 |
-
More information needed
|
| 34 |
|
| 35 |
-
|
| 36 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
|
| 38 |
-
Training
|
| 39 |
-
More information needed
|
| 40 |
|
| 41 |
-
Training
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
total_train_batch_size: 32
|
| 51 |
-
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 52 |
-
lr_scheduler_type: linear
|
| 53 |
-
lr_scheduler_warmup_steps: 500
|
| 54 |
-
num_epochs: 30
|
| 55 |
-
mixed_precision_training: Native AMP
|
| 56 |
-
Training results
|
| 57 |
-
Training Loss Epoch Step Validation Loss Wer
|
| 58 |
-
6.6905 3.84 400 0.7109 0.7800
|
| 59 |
-
0.4532 7.69 800 0.2972 0.3977
|
| 60 |
-
0.1957 11.54 1200 0.2907 0.3522
|
| 61 |
-
0.1284 15.38 1600 0.3117 0.3317
|
| 62 |
-
0.0979 19.23 2000 0.3000 0.3353
|
| 63 |
-
0.0749 23.08 2400 0.2823 0.3045
|
| 64 |
-
0.0584 26.92 2800 0.2982 0.2885
|
|
|
|
| 24 |
value: 0.2884547694473777
|
| 25 |
---
|
| 26 |
|
| 27 |
+
Telugu-ASR
|
| 28 |
+
This model is a fine-tuned version of Wav2Vec2 on the openslr dataset.
|
| 29 |
+
It achieves the following results on the evaluation set:
|
| 30 |
|
| 31 |
+
- Loss: 0.2982
|
| 32 |
+
- Wer: 0.2885
|
|
|
|
|
|
|
| 33 |
|
| 34 |
+
## Model description
|
| 35 |
+
- learning_rate: 0.0003
|
| 36 |
+
- train_batch_size: 16
|
| 37 |
+
- eval_batch_size: 8
|
| 38 |
+
- seed: 42
|
| 39 |
+
- gradient_accumulation_steps: 2
|
| 40 |
+
- total_train_batch_size: 32
|
| 41 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 42 |
+
- lr_scheduler_type: linear
|
| 43 |
+
- lr_scheduler_warmup_steps: 500
|
| 44 |
+
- num_epochs: 30
|
| 45 |
+
- mixed_precision_training: Native AMP
|
| 46 |
|
| 47 |
+
### Training results
|
|
|
|
| 48 |
|
| 49 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
| 50 |
+
|:-------------:|:-----:|:----:|:---------------:|:------:|
|
| 51 |
+
| 6.6905 | 3.84 | 400 | 0.7109 | 0.7800 |
|
| 52 |
+
| 0.4532 | 7.69 | 800 | 0.2972 | 0.3977 |
|
| 53 |
+
| 0.1957 | 11.54 | 1200 | 0.2907 | 0.3522 |
|
| 54 |
+
| 0.1284 | 15.38 | 1600 | 0.3117 | 0.3317 |
|
| 55 |
+
| 0.0979 | 19.23 | 2000 | 0.3000 | 0.3353 |
|
| 56 |
+
| 0.0749 | 23.08 | 2400 | 0.2823 | 0.3045 |
|
| 57 |
+
| 0.0584 | 26.92 | 2800 | 0.2982 | 0.2885 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|