{
  "type": "SpeechEncoderModelConfig",
  "vocab_size": 512,
  "sample_rate": 16000,
  "n_fft": 400,
  "n_mels": 100,
  "hop_length": 160,
  "d_model": 1024,
  "max_seq_len": 1024,
  "gradient_checkpointing": false,
  "n_query_groups": 16,
  "n_heads": 16,
  "use_qkv_bias": false,
  "positional_encoding_name": "rotary",
  "use_split_rope": false,
  "split_rope_local_axes": 1,
  "split_rope_scale": 7.5,
  "rotary_embedding_dim": 64,
  "attention_dropout_p": 0.0,
  "proj_dropout_p": 0.0,
  "is_causal": false,
  "pre_attn_norm_type": "rms",
  "pre_mlp_norm_type": "rms",
  "mlp_dropout_p": 0.0,
  "fsq_levels": [
    8,
    8,
    8,
    6,
    5
  ],
  "fsq_eps": 0.001,
  "subsampler_stride": 2,
  "n_encoder1_layers": 6,
  "n_premerge_layers": 3,
  "frame_merge": 2,
  "fsmn_kernel_size": 31,
  "merged_fsmn_kernel_size": 15,
  "n_encoder2_layers": 6,
  "ctc_upsample": 1,
  "n_spec_freq_masks": 2,
  "spec_freq_width": 27,
  "n_spec_time_masks": 10,
  "spec_time_fraction": 0.05,
  "enable_fsq_ctc_head": false,
  "weight_init": "scaled_normal",
  "init_std": 0.02
}