# visopt_audio_audiocaps_caption_reptok_6clst_l19_31_nocache_noeostk_ptl10_ptlr1e5fixed_ast_bs8_longt_lessmask_opt_2_7b
config=./configs/audio/ePALM_audio_caption.yaml
data_dir=data/audiocaps
output_dir=logs/epalm/ePALM_audio_caption

accelerate launch --multi_gpu --mixed_precision=fp16 --num_processes=4 --num_machines=1 accelerate_training/audio_caption.py \
--config $config \
--output_dir  $output_dir \
--data_dir $data_dir \
--save_best \
--text_model facebook/opt-2.7b \
--low_cpu

# --evaluate --checkpoint $WORK/logs/epalm/ePALM_audio_caption/checkpoint_best.pth