config.json

{
  "model_type": "audio-only-thinker",
  "architectures": ["AudioOnlyThinker"],
  "auto_map": {
    "AutoModelForCausalLM": "modeling_audio_only_thinker.AudioOnlyThinker"
  },
  "audio_token_index": 151646,
  "audio_start_token_id": 151647,
  "audio_end_token_id": 151648,
  "bos_token_id": 151644,
  "eos_token_id": 151645,
  "pad_token_id": 151643,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.50.0.dev0",
  "text_config": {
    "model_type": "qwen2_5_omni_text",
    "vocab_size": 152064,
    "hidden_size": 3584,
    "intermediate_size": 18944,
    "num_attention_heads": 28,
    "num_hidden_layers": 28,
    "num_key_value_heads": 4,
    "max_position_embeddings": 32768,
    "sliding_window": 32768,
    "max_window_layers": 28,
    "rope_theta": 1000000.0,
    "rope_scaling": {
      "mrope_section": [16, 24, 24],
      "type": "default",
      "rope_type": "default"
    },
    "use_cache": true,
    "use_sliding_window": false,
    "tie_word_embeddings": false,
    "init_std": 0.02,
    "rms_norm_eps": 1e-06,
    "hidden_act": "silu",
    "attention_dropout": 0.0
  },
  "audio_config": {
    "model_type": "qwen2_5_omni_audio_encoder",
    "d_model": 1280,
    "num_hidden_layers": 32,
    "encoder_attention_heads": 20,
    "encoder_ffn_dim": 5120,
    "dropout": 0.0,
    "activation_function": "gelu",
    "num_mel_bins": 128,
    "output_dim": 3584,
    "max_source_positions": 1500,
    "n_window": 100,
    "scale_embedding": false,
    "init_std": 0.02,
    "activation_dropout": 0.0,
    "attention_dropout": 0.0,
    "_attn_implementation_autoset": true
  }
}