Add the Core ML phone model (htdemucs_6s, 3.9 s, fp16 compute, float32 I/O) for the iOS app

Add the 3.9 s phone model with half-precision compute (for WebGPU devices with shader-f16)

Add config.json so downloads are counted (one request per device's first model download)

Add music analysis models (beats, chords, notes) and the demo song

Model card

Add HTDemucs 6-source ONNX exports (7.8 s default, 3.9 s phone)

initial commit