model:
  name: "openai/whisper-medium"
  language: "persian"
  task: "transcribe"

training:
  batch_size: 16
  learning_rate: 1e-5
  num_epochs: 10
  warmup_steps: 500
  gradient_accumulation_steps: 2
  save_steps: 1000
  eval_steps: 500
  logging_steps: 100

data:
  dataset_path: "confirmed_dataset/confirmed.parquet"
  max_audio_length: 30.0
  min_audio_length: 1.0
  train_split: 0.9
  eval_split: 0.1

persian:
  use_hazm: true
  normalize_text: true
  remove_diacritics: true

output:
  model_dir: "whisper-medium-finetuned/models"
  logs_dir: "whisper-medium-finetuned/logs"
  faster_whisper_dir: "whisper-medium-finetuned/faster_whisper"