config.json

{
  "model_type": "neurovoice",
  "architectures": [
    "NeuroVoiceModel"
  ],
  "auto_map": {
    "AutoConfig": "configuration_neurovoice.NeuroVoiceConfig",
    "AutoModel": "modeling_neurovoice.NeuroVoiceModel"
  },
  "torch_dtype": "bfloat16",
  "model_name": "NeuroVoice-0.5B",
  "version": "v0.1-alpha",
  "architecture": "TwoStageAudioLM",
  "text_backbone": "Qwen/Qwen2.5-0.5B",
  "audio_codec": "kyutai/mimi",
  "text_vocab_size": 151936,
  "audio_vocab_size": 2048,
  "num_codebooks": 8,
  "d_model": 896,
  "num_heads": 14,
  "num_kv_heads": 2,
  "num_layers": 24,
  "num_depth_layers": 4,
  "d_ff": 4864,
  "max_seq_len": 2048,
  "rope_theta": 1000000.0,
  "dropout_rate": 0.0,
  "activation": "swiglu",
  "dtype": "bfloat16",
  "use_qk_norm": false,
  "pad_token_id": 0,
  "bos_token_id": 1,
  "eos_token_id": 2,
  "unk_token_id": 3,
  "instruct_token_id": 4,
  "text_token_id": 5,
  "ref_audio_token_id": 6,
  "audio_start_token_id": 7,
  "audio_end_token_id": 8
}