{
  "vocoder": "attentionless vocoder",
  "llm_tts": "delayed streams modelling",
  "torch_dtype": "bfloat16",
  "architectures": [
    "LLMTransformer"
  ],
  "model_type": "moshi"
}