Loading custom models does not work, container expects the default models only

[TrackLab]

When running the base command

sudo docker run -it -p 10300:10300 --name test -v ~/whisperdata/:/data rhasspy/wyoming-whisper --model "base" --language de

The docker container boots up fine. However, I would like to use custom whisper models, such as german fine tuned ones: https://huggingface.co/primeline/whisper-large-v3-turbo-german

The V2 release of wyoming-faster-whisper states that this is easily possible, by passing the model ID "primeline/whisper-large-v3-turbo-german" into the --model parameter. Resulting in this command

sudo docker run -it -p 10300:10300 --name test -v ~/whisperdata/:/data rhasspy/wyoming-whisper --model "whisper-large-v3-turbo-german.de" --language de

However, this results in the error

Traceback (most recent call last):
  File "/usr/lib/python3.9/runpy.py", line 197, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.9/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/usr/local/lib/python3.9/dist-packages/wyoming_faster_whisper/__main__.py", line 155, in <module>
    run()
  File "/usr/local/lib/python3.9/dist-packages/wyoming_faster_whisper/__main__.py", line 150, in run
    asyncio.run(main())
  File "/usr/lib/python3.9/asyncio/runners.py", line 44, in run
    return loop.run_until_complete(main)
  File "/usr/lib/python3.9/asyncio/base_events.py", line 642, in run_until_complete
    return future.result()
  File "/usr/local/lib/python3.9/dist-packages/wyoming_faster_whisper/__main__.py", line 124, in main
    whisper_model = faster_whisper.WhisperModel(
  File "/usr/local/lib/python3.9/dist-packages/faster_whisper/transcribe.py", line 139, in __init__
    model_path = download_model(
  File "/usr/local/lib/python3.9/dist-packages/faster_whisper/utils.py", line 78, in download_model
    raise ValueError(
ValueError: Invalid model size 'whisper-large-v3-turbo-german.de', expected one of: tiny.en, tiny, base.en, base, small.en, small, medium.en, medium, large-v1, large-v2, large-v3, large, distil-large-v2, distil-medium.en, distil-small.en, distil-large-v3

A few times I even ran into a other error, stating
huggingface_hub.errors.HFValidationError: Repo id must use alphanumeric chars or '-', '_', '.', '--' and '..' are forbidden, '-' and '.' cannot start or end the name, max length is 96: 'primeline/whisper-tiny-german-1224 --language de'.

[kds69]

1. I am probably not fully literate on the syntax to be used, but why did you use:
   --model "whisper-large-v3-turbo-german.de"
   instead of
   --model primeline/whisper-large-v3-turbo-german

2. Anyway this model is failing on my end. I see it is missing at least model.bin on huggingface repository....

3. Custom models functionality do work for me.

[markusressel]

I tried using:

openai/whisper-base.en

just to test things out. It is present on HuggingFace: https://huggingface.co/openai/whisper-base.en
But starting the Docker image (v2.4.0) results in this error:

Unable to open file 'model.bin' in model '/data/models--openai--whisper-base/snapshots/e37978b90ca9030d5170a5c07aadb050351a65bb'

The "files" view on HuggingFace (see: https://huggingface.co/openai/whisper-base.en/tree/main) also shows that there is no such file in the repo.
Is there a way to specify the name of the file? I guess in that case pytorch_model.bin would be the correct file to use. Or would this then also just lead to the error mentioned by OP?

[markusressel]

I tested my theorie by creating my own repository, copying the mentioned pytorch_model.bin to my repo as model.bin.
See: https://huggingface.co/markus9100/whisper-base.en/tree/main

Now running it with markus9100/whisper-base.en results in this error:

Traceback (most recent call last):
  File "/usr/lib/python3.9/runpy.py", line 197, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.9/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/usr/local/lib/python3.9/dist-packages/wyoming_faster_whisper/__main__.py", line 155, in <module>
    run()
  File "/usr/local/lib/python3.9/dist-packages/wyoming_faster_whisper/__main__.py", line 150, in run
    asyncio.run(main())
  File "/usr/lib/python3.9/asyncio/runners.py", line 44, in run
    return loop.run_until_complete(main)
  File "/usr/lib/python3.9/asyncio/base_events.py", line 642, in run_until_complete
    return future.result()
  File "/usr/local/lib/python3.9/dist-packages/wyoming_faster_whisper/__main__.py", line 124, in main
    whisper_model = faster_whisper.WhisperModel(
  File "/usr/local/lib/python3.9/dist-packages/faster_whisper/transcribe.py", line 145, in __init__
    self.model = ctranslate2.models.Whisper(
RuntimeError: Unsupported model binary version. This executable supports models with binary version v6 or below, but the model has binary version v67324752. This usually means that the model was generated by a later version of CTranslate2. (Forward compatibility is not guaranteed.)

So I guess even the official base models aren't compatible with the container in this repo.

[markusressel]

With the new 2.5.0 container version at least the base image is now compatible, see:
#57 (comment)

The options are:

--use-transformers
--model openai/whisper-tiny.en
--language en

Not sure what --use-transformers does, but might be necessary for models that are not converted for use with faster-whisper 🤷

[synesthesiam]

Yes, --use-transformers is necessary for using non-faster-whisper models. Instead, it interprets --model as a HuggingFace transformers model id.