{
  "feature_dim": 198,
  "d_model": 256,
  "num_heads": 8,
  "num_layers": 6,
  "ffn_dim": 1024,
  "dropout": 0.1,
  "vocab_size": 8000,
  "max_seq_len": 5000,
  "time_mask_prob": 0.5,
  "time_mask_max_frames": 20,
  "frame_drop_prob": 0.05,
  "use_grad_checkpoint": true
}