MuQ-Jev

Candidate-based music classification with MuQ. Supports genre, mood, instruments, voice presence and vocal gender.

Usage

Install the inference code, then:

from muq_jev import MUQJEV

model = MUQJEV.from_pretrained("ABLingss/MUQ-JEV", device="cuda")
print(model.predict("music.wav", task="genre")["prediction"])

Up to 30 seconds per call. Scores are uncalibrated; custom-candidate performance may vary. Use policy="raw" for direct model scores. CPU is supported with device="cpu".

License

Inference code: MIT. Included fine-tuned MuQ weights retain CC-BY-NC-4.0. MiniLM weights/tokenizer retain Apache-2.0. See the included license files.