# visopt_vqa_vlt5_stok_6clst_l19_31_reptok_nocache_ptl10_ptlr1e5fixed_opt_2_7b config=./configs/image/ePALM_vqa.yaml data_dir=data/vl_adapter/vlt5_dataset output_dir=logs/epalm/ePALM_vqa accelerate launch --multi_gpu --mixed_precision=fp16 --num_processes=4 --num_machines=1 accelerate_training/vqa.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_vqa/checkpoint_last.pth \