Update README.md
Browse files
README.md
CHANGED
|
@@ -15,12 +15,22 @@ tags:
|
|
| 15 |
This model was fine-tuned on over 24 hours of transcribed upper sorbian speech to aid future research, conservation and revitalisation of the language.
|
| 16 |
|
| 17 |
|
| 18 |
-
## Training Data
|
| 19 |
- **Source:** Stiftung für das sorbische Volk / Załožba za serbski lud (https://stiftung.sorben.com/)
|
| 20 |
- **Volume:** 1493 Minutes, 10% Validation Set, 10% Test Set
|
| 21 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
## Performance
|
| 23 |
-
### Metrics
|
| 24 |
- **Word Error Rate:** 5.7
|
| 25 |
|
| 26 |
## Usage
|
|
|
|
| 15 |
This model was fine-tuned on over 24 hours of transcribed upper sorbian speech to aid future research, conservation and revitalisation of the language.
|
| 16 |
|
| 17 |
|
| 18 |
+
## Training Data
|
| 19 |
- **Source:** Stiftung für das sorbische Volk / Załožba za serbski lud (https://stiftung.sorben.com/)
|
| 20 |
- **Volume:** 1493 Minutes, 10% Validation Set, 10% Test Set
|
| 21 |
|
| 22 |
+
## Training Details
|
| 23 |
+
- **Hyperparameters**:
|
| 24 |
+
- Batch size: 64
|
| 25 |
+
- Learning rate: 3e-6
|
| 26 |
+
- Linear Decay
|
| 27 |
+
- **Optimizer**: AdamW
|
| 28 |
+
- **Warmup**: 1000 Steps
|
| 29 |
+
- **Additional Techniques**: BF16 Training, Frozen initial 15 layers
|
| 30 |
+
|
| 31 |
+
|
| 32 |
## Performance
|
| 33 |
+
### Metrics
|
| 34 |
- **Word Error Rate:** 5.7
|
| 35 |
|
| 36 |
## Usage
|