python run_distil_audio_multilabel_classification.py \
    --model_name_or_path MIT/ast-finetuned-audioset-10-10-0.4593 \
    --dataset_name bookbot/audioset \
    --output_dir distil-ast-audioset-2 \
    --overwrite_output_dir \
    --remove_unused_columns False \
    --freeze_feature_encoder False \
    --do_train --do_eval \
    --fp16 \
    --learning_rate 3e-5 \
    --alpha 0.5 \
    --temperature 2.0 \
    --layer_prefix audio_spectrogram_transformer.encoder.layer \
    --delimiter . \
    --teacher_blocks 0 2 4 6 8 10 \
    --attention_mask False \
    --warmup_ratio 0.1 \
    --num_train_epochs 10 \
    --per_device_train_batch_size 32 \
    --gradient_accumulation_steps 4 \
    --per_device_eval_batch_size 32 \
    --dataloader_num_workers 4 \
    --logging_strategy epoch \
    --evaluation_strategy epoch \
    --save_strategy epoch \
    --save_total_limit 3 \
    --seed 0 \
    --report_to tensorboard \
    --push_to_hub \
    --hub_model_id bookbot/distil-ast-audioset-2 \
    --hub_private_repo True \
    --use_auth_token True