|
--- |
|
license: apache-2.0 |
|
base_model: |
|
- openai/whisper-large-v3-turbo |
|
- sasikr2/whisper-large-v3-turbo |
|
pipeline_tag: automatic-speech-recognition |
|
tags: |
|
- ctranslate2 |
|
- whisper |
|
- 100 languages |
|
--- |
|
# Whisper Ctranslate2 |
|
Whisper large-v3-turbo is distilled version of large-v3, only difference is number of decoder layers. |
|
In turbo, there are only 4 decoder layers instead of 32 in large-v3. |
|
``` |
|
|
|
``` |
|
|
|
This repo contains the conversion of [whisper-large-v3-turbo openai checkpoint](https://github.com/openai/whisper) to [ctranslate2](https://github.com/OpenNMT/CTranslate2) format. |
|
|
|
Command to convert safetensors to ctranslate format checkpoint : |
|
```ct2-transformers-converter --model sasikr2/whisper-large-v3-turbo --output_dir whisper-large-v3-turbo-ct2 --quantization float16``` |
|
|
|
huggingface checkpoint: sasikr2/whisper-large-v3-turbo |
|
|
|
## Parent source repo |
|
[official github page](https://github.com/openai/whisper) |
