| ./run_t5_mlm_flax_streaming.py --output_dir="./norwegian-mt5" --model_name_or_path="./norwegian-mt5" --dataset_name="pere/nb_nn_balanced_shuffled" --max_seq_length="512" --per_device_train_batch_size="16" --learning_rate="1e-2" --weight_decay="0.001" --warmup_steps="5000" --overwrite_output_dir --num_train_epochs="5" --logging_steps="500" --save_steps="2500" --eval_steps="2500" --adafactor --push_to_hub --adafactor --preprocessing_num_workers 94 | |