training_config.json

{
  "transcript_file": "/home/metythorn/konai/services/speech-service/asr_training_data/asr_mixed_dataset.txt",
  "model_name": "openai/whisper-large-v3",
  "output_dir": "logs/whisper-large-v3-mixed-10eps-clean-text-199k",
  "num_train_epochs": 10,
  "learning_rate": 1e-06,
  "gradient_accumulation_steps": 1,
  "train_batch_size": 2,
  "eval_batch_size": 2,
  "warmup_steps": 1000,
  "weight_decay": 0.01,
  "max_label_length": 448,
  "max_audio_seconds": 30,
  "val_ratio": 0.05,
  "test_ratio": 0.01,
  "data_ratio": 1.0,
  "gradient_checkpointing": false,
  "lr_scheduler_type": "cosine",
  "dataset_cache_dir": "cache/raw_dataset",
  "force_rebuild_dataset": false,
  "seed": 42,
  "DEFAULT_VAL_RATIO": 0.05,
  "DEFAULT_TEST_RATIO": 0.01,
  "DEFAULT_DATA_RATIO": 1.0,
  "dataset_sample_rate": 16000,
  "dataset_version": 5,
  "dataset_config_filename": "config.json"
}