name: checkpoint
model_family: wav2vec2_llama
model_config:
  wav2vec2_asr_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: 9812
    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
  llama_config:
    model_dim: 4096
    max_seq_len: 8192
    vocab_size: 9812
    pad_idx: 1
    tied_embeddings: false
    num_layers: 12
    num_attn_heads: 8
    num_key_value_heads: 8
    ffn_inner_dim: 4096
    ffn_inner_dim_scale: 0.6666666666666666
    ffn_inner_dim_multiplier: 1.0
    ffn_inner_dim_multiple_of: 256
    rope_theta: 10000.0
    use_scaled_rope: false
    rope_scale:
      factor: 8.0
      frequency_factors:
      - 1.0
      - 4.0
      original_context_length: 8192
    dropout_p: 0.1
    init_std: null
    init_std_scale: layer
    shard_embed_dim: true
  beam_search_config:
    nbest: 5
    length_norm: false
    compression_window: 100
    compression_threshold: 4.0
  streaming_config:
    is_streaming: false
    segment_secs: 15.0
    sample_rate: 16000
    n_context_segments: 1
    text_tokenizer: ''
    min_audio_ms: 25
  encoder_stacking: 1
  frozen_encoder: 1
  lang_embeddings_p: 0.5
  language_column_name: lang
  context_text_only: false
  n_special_tokens: 1
  unk_idx: 3
  bos_idx: 0
  eos_idx: 2
  pad_idx: 1
  boh_idx: null
  eoh_idx: null
  model_type: LLM_ASR_LID
  n_context_examples: 0