Sample run:
python run_vit_gpt2.py \
--output_dir="testing" \
--data_dir="data/Flicker8k_Dataset" \
--train_file="data/train.tsv" \
--validation_file="data/val.tsv" \
--do_train --do_eval \
--num_train_epochs="3" --max_seq_length 256 \
--per_device_train_batch_size="64" \
--per_device_eval_batch_size="64" \
--learning_rate="5e-5" --warmup_steps="0" --weight_decay 0.1 \
--overwrite_output_dir \
--preprocessing_num_workers 32 \