config.json

{
  "audio_channels": 2,
  "bottom_channels": 512,
  "cac": true,
  "channels": 48,
  "channels_time": null,
  "context": 1,
  "context_enc": 0,
  "dconv_comp": 8,
  "dconv_depth": 2,
  "dconv_init": 0.001,
  "dconv_mode": 3,
  "depth": 4,
  "emb_scale": 10,
  "emb_smooth": true,
  "end_iters": 0,
  "freq_emb": 0.2,
  "growth": 2,
  "kernel_size": 8,
  "multi_freqs": [],
  "multi_freqs_depth": 3,
  "nfft": 4096,
  "norm_groups": 4,
  "norm_starts": 4,
  "rescale": 0.1,
  "rewrite": true,
  "samplerate": 44100,
  "segment": 7.8,
  "sources": [
    "drums",
    "bass",
    "other",
    "vocals"
  ],
  "stride": 4,
  "t_auto_sparsity": false,
  "t_cape_augment": true,
  "t_cape_glob_loc_scale": [
    5000.0,
    1.0,
    1.4
  ],
  "t_cape_mean_normalize": true,
  "t_cross_first": false,
  "t_dropout": 0.02,
  "t_emb": "sin",
  "t_gelu": true,
  "t_global_window": 100,
  "t_group_norm": false,
  "t_heads": 8,
  "t_hidden_scale": 4.0,
  "t_layer_scale": true,
  "t_layers": 5,
  "t_lr": null,
  "t_mask_random_seed": 42,
  "t_mask_type": "diag",
  "t_max_period": 10000.0,
  "t_max_positions": 10000,
  "t_norm_first": true,
  "t_norm_in": true,
  "t_norm_in_group": false,
  "t_norm_out": true,
  "t_sin_random_shift": 0,
  "t_sparse_attn_window": 400,
  "t_sparse_cross_attn": false,
  "t_sparse_self_attn": false,
  "t_sparsity": 0.95,
  "t_weight_decay": 0.0,
  "t_weight_pos_embed": 1.0,
  "time_stride": 2,
  "use_train_segment": true,
  "wiener_iters": 0,
  "wiener_residual": false
}