YvanM/MuQ-MuLan-large-onnx

MuQ-MuLan-large — ONNX

An ONNX export of OpenMuQ/MuQ-MuLan-large (revision 2e01c79), which embeds music and text in the same 512-dimensional space.

• audio.onnx — mono waveform at 24 kHz (float32, 10 s windows of 240 000 samples) → one L2-normalised 512-d vector per window. The mel front-end is inside the graph.
• text.onnx + tokenizer.json — text (up to 64 tokens) → an L2-normalised 512-d vector.
• manifest.json — inputs, outputs and file hashes.

Outputs match the original PyTorch model (cosine similarity ≥ 0.99999). Converted, not retrained.

License

CC BY-NC 4.0, like the original weights: non-commercial use only.