lora_config.yaml

# LoRA finetune related
use_adapter: true
adapter: lora
save_strategy: adapter
adapter_conf:
  rank: 4
  alpha: 4
  dropout_rate: 0.1
  target_modules:
    - w_1
    - w_2
    - merge_proj

freeze_param:
  - "encoder"
  - "ctc"

rir_scp: null
rir_apply_prob: 1.0
noise_scp: null
noise_apply_prob: 1.0
noise_db_range: '13_15'
speech_volume_normalize: null
non_linguistic_symbols: null

preprocessor_conf:
  speech_name: speech
  text_name: text

# training related
seed: 2024
num_workers: 1
ngpu: 1
batch_type: unsorted
batch_bins: 1600000
batch_size: 1
accum_grad: 4
max_epoch: 20
log_interval: 10
patience: null
init: null
best_model_criterion:
-   - valid
    - acc
    - max
keep_nbest_models: 10
use_amp: false
multiple_iterator: false

optim: adamw
optim_conf:
    lr: 0.0001
    weight_decay: 0.000001
scheduler: steplr
scheduler_conf:
    step_size: 1000

specaug: specaug
specaug_conf:
    apply_time_warp: true
    time_warp_window: 5
    time_warp_mode: bicubic
    apply_freq_mask: true
    freq_mask_width_range:
    - 0
    - 27
    num_freq_mask: 2
    apply_time_mask: true
    time_mask_width_ratio_range:
    - 0.
    - 0.05
    num_time_mask: 5