config.json

{
  "mulan": {
    "sr": 24000,
    "clip_secs": 10,
    "dim_latent": 512,
    "decoupled_contrastive_learning": true,
    "hierarchical_contrastive_loss": false,
    "hierarchical_contrastive_loss_layers": null,
    "sigmoid_contrastive_loss": false,
    "rank_contrast": true
  },
  "audio_model": {
    "name": "OpenMuQ/MuQ-large-msd-iter",
    "model_dim": 1024,
    "use_layer_idx": -1
  },
  "text_model": {
    "name": "xlm-roberta-base",
    "model_dim": null,
    "use_layer_idx": -1
  },
  "audio_transformer": {
    "dim": 768,
    "tf_depth": 0,
    "heads": 8,
    "dim_head": 64,
    "attn_dropout": 0,
    "ff_dropout": 0,
    "ff_mult": 4
  },
  "text_transformer": {
    "dim": 768,
    "tf_depth": 8,
    "max_seq_len": 1024,
    "dim_head": 64,
    "heads": 8,
    "attn_dropout": 0,
    "ff_dropout": 0,
    "ff_mult": 4
  }
}