Model assets for Pomegranate, a tiny speech model with a thinker-talker architecture.
Built using:
- Qwen/Qwen3-0.6B for Thinker and Talker backbones
- openai/whisper-large-v3 for speech encoder
- FunAudioLLM/CosyVoice2-0.5B for decoder assets
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support