config.json

{
  "architectures": [
    "Qwen3AudioWrappedForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "audio_adapter_configs": [
    {
      "adapter_embedding_dim": 1024,
      "adapter_name": "downsampler_conformer",
      "audio_encoder_layers": [
        7,
        15,
        23,
        31
      ],
      "downsampler_depth": 1,
      "encoder_embedding_dim": 1280,
      "encoder_name": "whisper",
      "layer_fusion_config": {
        "layer_fusion_type": "weighted_average"
      },
      "llm_embedding_dim": 2560,
      "norm_type": "batch",
      "pre_average": true,
      "use_conformer": false
    },
    {
      "adapter_embedding_dim": 1024,
      "adapter_name": "identity",
      "audio_encoder_layers": [
        3,
        7,
        11
      ],
      "encoder_name": "sslam",
      "layer_fusion_config": {
        "layer_fusion_type": "weighted_average"
      },
      "llm_embedding_dim": 2560,
      "pre_average": true,
      "use_llm_proj": false
    },
    {
      "adapter_embedding_dim": 1024,
      "adapter_name": "identity",
      "audio_encoder_layers": [
        4,
        8,
        12
      ],
      "encoder_name": "muq",
      "layer_fusion_config": {
        "layer_fusion_type": "weighted_average"
      },
      "llm_embedding_dim": 2560,
      "pre_average": true,
      "use_llm_proj": false
    },
    {
      "adapter_embedding_dim": 1024,
      "adapter_name": "identity",
      "audio_encoder_layers": [
        6,
        10,
        15,
        20,
        24
      ],
      "encoder_name": "w2vbert",
      "layer_fusion_config": {
        "layer_fusion_type": "weighted_average"
      },
      "llm_embedding_dim": 2560,
      "pre_average": true,
      "use_llm_proj": false
    }
  ],
  "audio_encoder_configs": [
    {
      "encoder_name": "whisper"
    },
    {
      "encoder_name": "sslam"
    },
    {
      "encoder_name": "muq"
    },
    {
      "encoder_name": "w2vbert"
    }
  ],
  "audio_fusion_config": {
    "conditional_embedding_dim": [
      1024,
      1024,
      768
    ],
    "conditional_encoders": [
      "w2vbert",
      "muq",
      "sslam"
    ],
    "first_cross_attention_layer_shared": [
      true,
      true,
      true
    ],
    "first_self_attention_block_shared": [
      true,
      true,
      true
    ],
    "fusion_type": "multiperceiver",
    "llm_embedding_dim": 2560,
    "main_encoder": "whisper",
    "num_cross_attention_heads": [
      4,
      4,
      4
    ],
    "num_cross_attention_layers": [
      1,
      1,
      1
    ],
    "num_latent_channels": [
      768,
      768,
      768
    ],
    "num_latents": [
      20,
      20,
      20
    ],
    "num_self_attention_blocks": [
      3,
      3,
      3
    ],
    "num_self_attention_heads": [
      4,
      4,
      4
    ],
    "num_self_attention_layers_per_block": [
      6,
      6,
      6
    ]
  },
  "audio_postprocessing_config": {
    "postprocessing_type": "identity"
  },
  "audio_sep_d_embed": 2560,
  "bos_token_id": 151643,
  "dtype": "float32",
  "eos_token_id": 151645,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2560,
  "initializer_range": 0.02,
  "intermediate_size": 9728,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 262144,
  "max_window_layers": 36,
  "model_type": "qwen3_audio",
  "num_attention_heads": 32,
  "num_hidden_layers": 36,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 5000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "transformers_version": "4.57.1",
  "use_cache": false,
  "use_explicit_audio_tokens": false,
  "use_sliding_window": false,
  "vocab_size": 151936
}