config.json

{
  "architectures": [
    "Qwen3OmniMoeThinkerForConditionalGeneration"
  ],
  "audio_config": {
    "activation_dropout": 0.0,
    "activation_function": "gelu",
    "attention_dropout": 0.0,
    "conv_chunksize": 500,
    "d_model": 32,
    "downsample_hidden_size": 32,
    "dropout": 0.0,
    "encoder_attention_heads": 4,
    "encoder_ffn_dim": 32,
    "encoder_layers": 2,
    "initializer_range": 0.02,
    "max_source_positions": 1500,
    "model_type": "qwen3_omni_moe_audio_encoder",
    "n_window": 50,
    "n_window_infer": 100,
    "num_mel_bins": 20,
    "output_dim": 32,
    "scale_embedding": false
  },
  "audio_end_token_id": 5,
  "audio_start_token_id": 4,
  "audio_token_id": 1,
  "dtype": "float32",
  "image_token_id": 2,
  "initializer_range": 0.02,
  "model_type": "qwen3_omni_moe_thinker",
  "position_id_per_seconds": 13,
  "seconds_per_chunk": 2,
  "text_config": {
    "attention_bias": false,
    "attention_dropout": 0.0,
    "bos_token_id": null,
    "decoder_sparse_step": 1,
    "eos_token_id": null,
    "hidden_act": "silu",
    "hidden_size": 32,
    "initializer_range": 0.02,
    "intermediate_size": 37,
    "max_position_embeddings": 1024,
    "max_window_layers": 3,
    "mlp_only_layers": [],
    "model_type": "qwen3_omni_moe_text",
    "moe_intermediate_size": 32,
    "norm_topk_prob": true,
    "num_attention_heads": 4,
    "num_experts": 8,
    "num_experts_per_tok": 2,
    "num_hidden_layers": 4,
    "num_key_value_heads": 2,
    "output_router_logits": false,
    "pad_token_id": 151643,
    "rms_norm_eps": 1e-06,
    "rope_parameters": {
      "interleaved": true,
      "mrope_section": [
        1,
        1,
        2
      ],
      "rope_theta": 1000000.0,
      "rope_type": "default",
      "type": "default"
    },
    "router_aux_loss_coef": 0.001,
    "sliding_window": 50,
    "tie_word_embeddings": false,
    "use_cache": true,
    "use_sliding_window": false,
    "vocab_size": 151676
  },
  "tie_word_embeddings": false,
  "transformers_version": "5.16.0.dev0",
  "user_token_id": 6,
  "video_token_id": 3,
  "vision_config": {
    "deepstack_visual_indexes": [
      1
    ],
    "depth": 2,
    "embed_dim": 32,
    "hidden_act": "quick_gelu",
    "hidden_size": 32,
    "in_channels": 3,
    "initializer_range": 0.02,
    "intermediate_size": 24,
    "mlp_ratio": 4,
    "model_type": "qwen3_omni_moe_vision_encoder",
    "num_heads": 4,
    "num_position_embeddings": 2304,
    "out_hidden_size": 32,
    "patch_size": 16,
    "spatial_merge_size": 1,
    "temporal_patch_size": 2
  },
  "vision_end_token_id": 8,
  "vision_start_token_id": 7
}