{
  "model_type": "xlsr_mamba",
  "architecture": "XLS-R + Bidirectional Mamba State-Space Audio Deepfake Detector",
  "frontend": {
    "type": "wav2vec2_xlsr",
    "name": "xlsr_300m",
    "sample_rate": 16000,
    "feat_dim": 1024,
    "encoder_layers": 24,
    "encoder_embed_dim": 1024,
    "encoder_num_heads": 16,
    "encoder_ff_interm_features": 4096,
    "conv_layer_config": [
      [512, 10, 5],
      [512, 3, 2],
      [512, 3, 2],
      [512, 3, 2],
      [512, 3, 2],
      [512, 2, 2],
      [512, 2, 2]
    ]
  },
  "projection": {
    "in_features": 1024,
    "out_features": 144,
    "bias": true
  },
  "first_bn": {
    "num_features": 1
  },
  "backbone": {
    "type": "BiMambaConformer",
    "d_model": 144,
    "num_layers": 6,
    "d_state": 16,
    "d_conv": 4,
    "expand": 2,
    "d_inner": 288,
    "dt_rank": 9,
    "pooling": "bidirectional_self_attention",
    "classifier": {
      "in_features": 144,
      "num_classes": 2
    }
  }
}