Add W8A16-multifunction MultiCodeDecoder assets (0.6B + 1.7B)

#2

Adds a new W8A16-multifunction/ MultiCodeDecoder variant for both Qwen3-TTS sizes (12hz-0.6b-customvoice and 12hz-1.7b-customvoice).

  • Single .mlmodelc carrying both the stepped (default) and fused functions, selected at load time via MLModelConfiguration.functionName (requires macOS 15 / iOS 18). Mirrors the multifunction SpeechDecoder rollout.
  • Additive: the existing single-function W8A16/ assets are unchanged and stay published for older SDKs.
EduardoPacheco changed pull request status to merged

Sign up or log in to comment