{
  "architectures": [
    "Phi4MultimodalModel"
  ],
  "attention_dropout": 0.0,
  "audio_config": {
    "activation": "swish",
    "audio_token_id": 200011,
    "bias_max_distance": 1000,
    "bias_symmetric": false,
    "chunk_size": -1,
    "conv_activation": "swish",
    "conv_glu_type": "swish",
    "depthwise_multiplier": 1,
    "depthwise_separable_out_channel": 128,
    "downsample_rate": 1,
    "dropout_rate": 0.0,
    "ext_pw_out_channel": 1024,
    "feature_layer": -2,
    "hidden_size": 32,
    "initializer_range": 1e-05,
    "input_size": 80,
    "intermediate_size": 48,
    "kernel_size": 3,
    "left_chunk": 18,
    "model_type": "phi4_multimodal_audio",
    "nemo_activation": "relu",
    "nemo_conv_channels": 128,
    "nemo_final_size": 10,
    "num_attention_heads": 8,
    "num_blocks": 2,
    "time_reduction": 8
  },
  "bos_token_id": 199999,
  "dtype": "float32",
  "embd_pdrop": 0.0,
  "eos_token_id": 199999,
  "hidden_act": "silu",
  "hidden_size": 32,
  "initializer_range": 0.02,
  "intermediate_size": 64,
  "max_position_embeddings": 4096,
  "model_type": "phi4_multimodal",
  "num_attention_heads": 4,
  "num_hidden_layers": 2,
  "num_key_value_heads": 2,
  "original_max_position_embeddings": 4096,
  "pad_token_id": 199999,
  "resid_pdrop": 0.0,
  "rms_norm_eps": 1e-05,
  "rope_parameters": {
    "partial_rotary_factor": 1.0,
    "rope_theta": 10000.0,
    "rope_type": "default"
  },
  "sliding_window": null,
  "tie_word_embeddings": false,
  "transformers_version": "5.16.0.dev0",
  "use_cache": true,
  "vision_config": {
    "attention_dropout": 0.0,
    "crop_size": 16,
    "feature_layer": -2,
    "hidden_act": "gelu_pytorch_tanh",
    "hidden_size": 32,
    "image_size": 448,
    "image_token_id": 200010,
    "initializer_range": 1e-05,
    "intermediate_size": 64,
    "layer_norm_eps": 1e-06,
    "model_type": "phi4_multimodal_vision",
    "num_attention_heads": 8,
    "num_channels": 3,
    "num_hidden_layers": 2,
    "patch_size": 14
  },
  "vocab_size": 200029
}