Added some description for Whisper.cpp
Browse files
README.md
CHANGED
|
@@ -146,6 +146,9 @@ model.generate(**inputs, use_cache=True, assistant_model=assistant_model)
|
|
| 146 |
In the original `simonl0909/whisper-large-v2-cantonese` model, it runs at 0.714s/sample for a CER of 7.65. \
|
| 147 |
Using speculative decoding with `alvanlii/whisper-small-cantonese`, it runs at 0.137s/sample for a CER of 7.67, which is much faster.
|
| 148 |
|
|
|
|
|
|
|
|
|
|
| 149 |
## Training Hyperparameters
|
| 150 |
- learning_rate: 5e-5
|
| 151 |
- train_batch_size: 25 (on 1 3090 GPU)
|
|
|
|
| 146 |
In the original `simonl0909/whisper-large-v2-cantonese` model, it runs at 0.714s/sample for a CER of 7.65. \
|
| 147 |
Using speculative decoding with `alvanlii/whisper-small-cantonese`, it runs at 0.137s/sample for a CER of 7.67, which is much faster.
|
| 148 |
|
| 149 |
+
## Whisper.cpp
|
| 150 |
+
Uploaded a GGML bin file for Whisper cpp as of June 2024. You can download the bin file [here](https://huggingface.co/alvanlii/whisper-small-cantonese/blob/main/ggml-model.bin) and try it out [here](https://whisper.ggerganov.com/).
|
| 151 |
+
|
| 152 |
## Training Hyperparameters
|
| 153 |
- learning_rate: 5e-5
|
| 154 |
- train_batch_size: 25 (on 1 3090 GPU)
|