run_distillation.json

{
  "model": {
    "teacher_model_path_or_name": "openai/whisper-base",
    "student_model_path": "_student_whisper",
    "processor_path_or_name": "openai/whisper-base",
    "distil_model_path": "_distil_whisper"
  },
  "data": {
    "train_jsonl_path": "pseudo_labeled_dataset.jsonl",
    "dev_jsonl_path": "./demo_data/demo_jsonl_dataset.jsonl",
    "test_jsonl_path": "./demo_data/demo_jsonl_dataset.jsonl",
    "audio_duration_col": "input_length",
    "audio_path_col": "path", 
    "text_col": "text", 
    "metric_col": "cer/wer",
    "min_audio_duration": 0,
    "max_audio_duration": 30, 
    "min_token_num": 0,
    "max_token_num": 448,
    "min_pseudo_label_metric": 0.2
  },
  "train": {
    "epochs": 10,
    "batch_size": 16,
    "learning_rate": 0.00001,
    "teacher_model_device": "cuda:0",
    "student_model_device": "cuda:3",
    "lr_decay_gamma": 0.9
  },
  "common": {
    "lang": "mandarin",
    "sampling_rate": 16000
  }
}