Not all models work after download

[theKHR3b]

Using docker version speaches:

speaches:
    image: ghcr.io/speaches-ai/speaches:latest-cuda
    depends_on:
      - openwebui
    ports:
      - "8000:8000"
    volumes:
      - speaches-hf-hub-cache:/home/ubuntu/.cache/huggingface/hub
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

volumes:
  speaches-hf-hub-cache:
    external: true

curl http://speaches:8000/v1/registry
gives following result:
...... many lines ommited {"id":"ytrbqrkflbvbhy/whisper-podlodka-turbo-ct2-float16","created":1765519359,"object":"model","owned_by":"ytrbqrkflbvbhy","language":["ru","en"],"task":"automatic-speech-recognition"}],"object":"list"}

This models is fine for me, try to download:
curl http://speaches:8000/v1/models/ytrbqrkflbvbhy/whisper-podlodka-turbo-ct2-float16 -X POST
Model 'ytrbqrkflbvbhy/whisper-podlodka-turbo-ct2-float16' downloaded

Ok, try to look at models list:
curl http://speaches:8000/v1/models
{"data":[],"object":"list"}
its empty!

Same happends with another model named juasker/whisper-ct2-podlodka-turbo

But if we try to load Systran/faster-distil-whisper-small.en like in documentation, it work as expected.
curl http://speaches:8000/v1/models
{"data":[{"id":"Systran/faster-distil-whisper-small.en","created":1765776806,"object":"model","owned_by":"Systran","language":["en"],"task":"automatic-speech-recognition"}]

Maybe there some problems with type of model, or naming, or something else.

[autronix]

I'm experiencing the same issue.
I ran curl -X POST http://127.0.0.1:8000/v1/models/sorendal/skyrim-whisper-medium.en , it says Downloaded and when I run the models listing endpoint, I only see whisper large (which I had downloaded earlier).

The model appears in the registry:

 {
      "id": "sorendal/skyrim-whisper-medium.en",
      "created": 1743541946,
      "object": "model",
      "owned_by": "sorendal",
      "language": [
        "en"
      ],
      "task": "automatic-speech-recognition"
    }

[ViViDboarder]

I'm experiencing the same with deepdml/whisper-large-v3-turbo, which also shows in the registry. No errors in the log either. The speaches server returns a 200 for the POST request.

[nabbl]

Did you guys fix this at some point ?

[vaaale]

I had some similar issues. There are some strange filterings going on in the list_local_models method. I have created a fork where I fixed these to make it work for me at least. You are welcome to try it.
https://github.com/vaaale/speaches.git