config/ear_vae_v2.json

{
  "transformer": {
    "depth": 2,
    "config": {
      "rotary_pos_emb": true,
      "dim_heads": 32
    }
  },
  "encoder": {
    "config": {
      "in_channels": 2,
      "channels": 128,
      "c_mults": [1, 2, 4, 8, 16],
      "strides": [2, 2, 4, 6, 10],
      "latent_dim": 128,
      "use_snake": true
    }
  },
  "decoder": {
    "config": {
      "out_channels": 2,
      "channels": 128,
      "c_mults": [1, 2, 4, 8, 16],
      "strides": [2, 2, 4, 6, 10],
      "latent_dim": 64,
      "use_nearest_upsample": false,
      "use_snake": true,
      "final_tanh": false
    }
  },
  "latent_dim": 64,
  "downsampling_ratio": 960,
  "io_channels": 2
}