python run_whisper_finetuning.py \ --model_name_or_path="openai/whisper-small" \ --output_dir="../whisper-testrun1" \ --repo_id="NbAiLab/whisper-testrun1" \ --overwrite_output_dir=True \ --language="Norwegian" \ --task="transcribe" \ --dataset_name="NbAiLab/NPSC" \ --dataset_config="16K_mp3" \ --do_train=True \ --do_eval=True \ --audio_column_name="audio" \ --text_column_name="normsentence_text" \ --per_device_train_batch_size=16 \ --per_device_train_batch_size=16 \ --learning_rate=2e-5 \ --warmup_steps=500 \ --max_steps=10000 \ --gradient_checkpointing=True \ --gradient_accumulation_steps=1 \ --group_by_length=False \ --evaluation_strategy="steps" \ --save_steps=1000 \ --eval_steps=1000 \ --logging_steps=250 \ --fp16=True \ --load_best_model_at_end=True \ --metric_for_best_model="wer" \ --greater_is_better=False \ --report_to="tensorboard" \ --predict_with_generate=True \ --generation_max_length=225 \ --print_training_arguments=True \ --push_to_hub=True