config.json

{
  "model_type": "zipformer-transducer",
  "architectures": ["ZipformerRnntModel"],
  "language": "vi",
  "sample_rate": 16000,
  "vocab_size": 3000,
  "tokens": "tokens.txt",
  "bpe_model": "bpe.model",
  "pytorch": {
    "checkpoint": "best-valid-loss.pt",
    "library": "icefall",
    "use_model_avg": true
  },
  "base_model": "actableai/zipformer-rnnt-v2",
  "finetune_dataset": "actableai/vi-meeting-soniox-wer16",
  "encoder_dim": "192,256,384,512,384,256",
  "num_encoder_layers": "2,2,3,4,3,2",
  "downsampling_factor": "1,2,4,8,4,2",
  "feedforward_dim": "512,768,1024,1536,1024,768",
  "num_heads": "4,4,4,8,4,4",
  "encoder_unmasked_dim": "192,192,256,256,256,192",
  "cnn_module_kernel": "31,31,15,15,15,31",
  "decoder_dim": 512,
  "joiner_dim": 512,
  "use_transducer": true,
  "use_ctc": false,
  "feature_dim": 80,
  "blank_id": 0,
  "context_size": 2
}