Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model:
|
| 3 |
+
- openai/whisper-large-v3
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# Model Description
|
| 7 |
+
This model is based on Whisper Large and was fine-tuned of transcribed upper sorbian speech. This is intended to contribute to the preservation of the Upper Sorbian language and can be freely used for language preservation projects in accordance with the license.
|
| 8 |
+
|
| 9 |
+
Note: This model is a snapshot from an ongoing fine-tuning session. Further (improved) versions of this will follow shortly.
|
| 10 |
+
|
| 11 |
+
# Model Details
|
| 12 |
+
- Model Name: DL-whisper-large-hsb
|
| 13 |
+
- Model Version: 1.0.0
|
| 14 |
+
- Model Date: 2025-03-05
|
| 15 |
+
- Publisher:
|
| 16 |
+
- Architecture: Whisper Large
|
| 17 |
+
- License:
|
| 18 |
+
|
| 19 |
+
# Training Details
|
| 20 |
+
Parameters:
|
| 21 |
+
|
| 22 |
+
- Learning rate: 6.25 x 10-⁶, linear decay
|
| 23 |
+
- Warmup: 1500 steps
|
| 24 |
+
- Additional: FP16 training
|
| 25 |
+
- Train batch size: 8
|
| 26 |
+
- Gradient accumulation step: 1
|
| 27 |
+
- Performance
|
| 28 |
+
|
| 29 |
+
Word Error Rate: 5.453
|