name: checkpoint
model_family: wav2vec2_asr
model_config:
  encoder_config:
    model_dim: 1024
    max_seq_len: 4096
    feature_dim: 512
    use_fbank: false
    first_pass_dropout_p: 0.0
    layer_norm_features: false
    feature_extractor_layer_descs:
    - - 512
      - 10
      - 5
    - - 512
      - 3
      - 2
    - - 512
      - 3
      - 2
    - - 512
      - 3
      - 2
    - - 512
      - 3
      - 2
    - - 512
      - 2
      - 2
    - - 512
      - 2
      - 2
    feature_extractor_bias: true
    feature_extractor_layer_norm_convs: true
    feature_grad_scale: 0.1
    num_fbank_channels: 0
    fbank_stride: 0
    sample_fbank_every_k: 0
    pos_encoder_type: conv
    pos_encoder_depth: 1
    pos_conv_kernel_size: 128
    num_pos_conv_groups: 16
    use_conformer: false
    num_encoder_layers: 24
    num_encoder_attn_heads: 16
    ffn_inner_dim: 4096
    dropout_p: 0.0
    attn_dropout_p: 0.0
    ffn_inner_dropout_p: 0.1
    layer_drop_p: 0.1
    norm_order: PRE
    depthwise_conv_kernel_size: 0
  target_vocab_size: 10288
  final_dropout_p: 0.0
  use_masking: false
  temporal_mask_span_len: 10
  max_temporal_mask_prob: 0.0
  min_num_temporal_mask_spans: 2
  spatial_mask_span_len: 64
  max_spatial_mask_prob: 0.0
  min_num_spatial_mask_spans: 2