python run_audio_classification.py \ --model_name_or_path openai/whisper-base \ --dataset_name common_language \ --output_dir whisper-base-ft-common-language-id \ --overwrite_output_dir \ --remove_unused_columns False \ --do_train \ --do_eval \ --fp16 \ --learning_rate 1e-5 \ --max_length_seconds 30 \ --attention_mask False \ --warmup_ratio 0.1 \ --num_train_epochs 10 \ --per_device_train_batch_size 32 \ --per_device_eval_batch_size 16 \ --gradient_checkpointing \ --dataloader_num_workers 4 \ --logging_strategy steps \ --logging_steps 25 \ --evaluation_strategy epoch \ --save_strategy epoch \ --load_best_model_at_end True \ --metric_for_best_model accuracy \ --seed 0 \ --freeze_feature_encoder False \ --label_column_name language \ --push_to_hub