config.json

{
  "d_model": 256,
  "n_heads": 4,
  "ffn_dim": 512,
  "n_layers": 6,
  "vocab_size": 10000,
  "patch_size": 16,
  "max_seq_len": 512,
  "dropout": 0.1,
  "audio_feat_dim": 768
}