How to use https://huggingface.co/openai/whisper-large-v3-turbo with whisper.cpp?

[raphael10-collab]

How to use this safetensors model : https://huggingface.co/openai/whisper-large-v3-turbo with whisper.cpp?

I tried, without success to convert mode.safetensors to ggml with llama.cpp and got "Model not supported".
And following the indications here: https://gist.github.com/AltAcc1812/0bc4b80838b49b6c3bdce9d94b66cb54 I got :

(.venv) raphy@raohy:~/whisper.cpp/scripts$ ./convert-all.sh ../models/whisper-large-v3-turbo/model.safetensors --outtype f16 --outfile ../models/whisper-large-v3-turbo.ggml
python3: can't open file '/home/raphy/whisper.cpp/scripts/models/convert-pt-to-ggml.py': [Errno 2] No such file or directory
mv: cannot stat 'models/ggml-model.bin': No such file or directory

[Nindaleth]

You can get it already converted like this: ./models/download-ggml-model.sh large-v3-turbo
Which downloads it from https://huggingface.co/ggerganov/whisper.cpp

Originally I found that you can call make -j large-v3-turbo as described in the README, the rest I found through the logging.

[raphael10-collab]

Hi @Nindaleth

raphy@raohy:~/whisper.cpp/models$ ./download-ggml-model.sh large-v3-turbo
Invalid model: large-v3-turbo

Available models:
  tiny tiny.en tiny-q5_1 tiny.en-q5_1
  base base.en base-q5_1 base.en-q5_1
  small small.en small.en-tdrz small-q5_1 small.en-q5_1
  medium medium.en medium-q5_0 medium.en-q5_0
  large-v1 large-v2 large-v2-q5_0 large-v3 large-v3-q5_0

But I've found that ggml-models available can be downloaded here: https://huggingface.co/ggerganov/whisper.cpp/tree/main

Thank you

[Nindaleth]

I'm quite sure that large-v3-turbo is supported as well: https://github.com/ggerganov/whisper.cpp/blob/master/models/download-ggml-model.sh#L55-L57
Maybe you didn't pull the latest state of whisper.cpp at that time?

Happy you solved your issue!

[raphael10-collab]

Yes. May be it was an "old" whisper.cpp version (three weeks ago). Git cloned the newest whisper.ccp and for the ggml-large-v3-turbo as well :

raphy@raohy:~/whisper.cpp/models$ ./download-ggml-model.sh large-v3-turbo
Downloading ggml model large-v3-turbo from 'https://huggingface.co/ggerganov/whisper.cpp' ...
ggml-large-v3-turbo.bin                            100%[================================================================================================================>]   1,51G   109MB/s    in 14s     
Done! Model 'large-v3-turbo' saved in '/home/raphy/whisper.cpp/models/ggml-large-v3-turbo.bin'
You can now use it like this:

  $ ./build/bin/whisper-cli -m /home/raphy/whisper.cpp/models/ggml-large-v3-turbo.bin -f samples/jfk.wav

Thank you