| # 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 \ | |