[BUG] Can't load the Qwen2 Audio Instruct model

[yovelcohen]

Issue Description

I'm trying to load the Qwen2 audio instruct model using the python interface.
It downloads the model successfully but than runs into a huggingface issue.

Steps to Reproduce

model = NexaVoiceInference("Qwen2-Audio-7.8B-Instruct:q6_K")

Stdout:

Downloading Qwen2-Audio-7.8B-Instruct/model-q6_K.gguf...
model-q6_K.gguf: 100%|██████████| 5.93G/5.93G [36:29<00:00, 2.91MB/s]
Successfully downloaded Qwen2-Audio-7.8B-Instruct/model-q6_K.gguf to /Users/user/.cache/nexa/hub/official/Qwen2-Audio-7.8B-Instruct/model-q6_K.gguf
Successfully pulled model Qwen2-Audio-7.8B-Instruct:model-q6_K to /Users/user/.cache/nexa/hub/official/Qwen2-Audio-7.8B-Instruct/model-q6_K.gguf, run_type: AudioLM

After the download's done, I get the following error:

    model = NexaVoiceInference("Qwen2-Audio-7.8B-Instruct:q6_K")
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/user/PycharmProjects/glixtranslationsdurable/.venv/lib/python3.11/site-packages/nexa/gguf/nexa_inference_voice.py", line 64, in __init__
    self._load_model()
  File "/Users/user/PycharmProjects/glixtranslationsdurable/.venv/lib/python3.11/site-packages/nexa/utils.py", line 300, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/user/PycharmProjects/glixtranslationsdurable/.venv/lib/python3.11/site-packages/nexa/gguf/nexa_inference_voice.py", line 76, in _load_model
    self.model = WhisperModel(
                 ^^^^^^^^^^^^^
  File "/Users/user/PycharmProjects/glixtranslationsdurable/.venv/lib/python3.11/site-packages/faster_whisper/transcribe.py", line 628, in __init__
    model_path = download_model(
                 ^^^^^^^^^^^^^^^
  File "/Users/user/PycharmProjects/glixtranslationsdurable/.venv/lib/python3.11/site-packages/faster_whisper/utils.py", line 107, in download_model
    return huggingface_hub.snapshot_download(repo_id, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/user/PycharmProjects/glixtranslationsdurable/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_validators.py", line 106, in _inner_fn
    validate_repo_id(arg_value)
  File "/Users/user/PycharmProjects/glixtranslationsdurable/.venv/lib/python3.11/site-packages/huggingface_hub/utils/_validators.py", line 154, in validate_repo_id
    raise HFValidationError(
huggingface_hub.errors.HFValidationError: Repo id must be in the form 'repo_name' or 'namespace/repo_name': '/Users/user/.cache/nexa/hub/official/Qwen2-Audio-7.8B-Instruct/model-q6_K.gguf'. Use `repo_type` argument if needed.

OS

macOS Sequoia

Python Version

3.11

Nexa SDK Version

0.0.9.4

GPU (if using one)

No response

[Davidqian123]

python interface of qwen2 audio is not supported for now, but in our roadmap.