{
  "model_type": "midashenglm-gen",
  "architectures": [
    "MiDashengLMGenForAudioGeneration"
  ],
  "auto_map": {
    "AutoConfig": "configuration_midashenglm.MiDashengLMGenConfig",
    "AutoModel": "modeling_midashenglm.MiDashengLMGenForAudioGeneration"
  },
  "seq_len": 500,
  "audio_emb_size": 768,
  "llm_emb_dim": 2048,
  "target_emb_dim": 768,
  "flow_dit_width": 2048,
  "flow_dit_depth": 16,
  "flow_dit_heads": 8,
  "flow_dit_mlp_ratio": 4.0,
  "cfg_trainable": true,
  "llm_model": "Qwen/Qwen3-1.7B",
  "eval_cfg": 2.0,
  "stop_threshold": 0.5,
  "min_stop_step": 5,
  "sample_rate": 16000,
  "torch_dtype": "float32"
}