{
  "aff_block_r": 4,
  "architectures": [
    "ClapAudioModelWithProjection"
  ],
  "attention_dropout": 0.1,
  "attention_probs_dropout_prob": 0.0,
  "depths": [
    2,
    2
  ],
  "drop_path_rate": 0.0,
  "dropout": 0.1,
  "dtype": "float32",
  "enable_fusion": false,
  "enable_patch_layer_norm": true,
  "flatten_patch_embeds": true,
  "freq_ratio": 2,
  "fusion_type": null,
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.1,
  "hidden_size": 32,
  "image_size": 60,
  "initializer_factor": 1.0,
  "initializer_range": 0.02,
  "intermediate_size": 37,
  "layer_norm_eps": 1e-05,
  "mlp_ratio": 4.0,
  "model_type": "clap_audio_model",
  "num_attention_heads": [
    2,
    2
  ],
  "num_channels": 3,
  "num_classes": 527,
  "num_hidden_layers": 2,
  "num_mel_bins": 16,
  "patch_embed_input_channels": 1,
  "patch_embeds_hidden_size": 16,
  "patch_size": 2,
  "patch_stride": 2,
  "projection_dim": 32,
  "projection_hidden_act": "relu",
  "qkv_bias": true,
  "spec_size": 64,
  "transformers_version": "5.15.0.dev0",
  "window_size": 4
}