export TRANSFORMERS_CACHE=/mnt/gpu_data1/kubapok/cache
python run_t5_mlm_flax.py \
	--output_dir="./temporal-t5-base" \
	--model_type="t5" \
	--config_name="./temporal-t5-base" \
	--tokenizer_name="./temporal-t5-base" \
	--train_file="./train-splitted-shuf.txt" \
	--validation_file="./dev-splitted-shuf.txt" \
	--max_seq_length="512" \
	--per_device_train_batch_size="32" \
	--per_device_eval_batch_size="32" \
	--adafactor \
	--learning_rate="0.005" \
	--weight_decay="0.001" \
	--warmup_steps="2000" \
	--overwrite_output_dir \
	--logging_steps="500" \
	--save_steps="10000" \
	--eval_steps="2500" \