python run_summarization.py \ --model_name_or_path google/t5-v1_1-base \ --do_train \ --train_file ../train/huggingface_format_year_as_text.csv \ --validation_file ../dev-0/huggingface_format_year_as_text.csv \ --source_prefix "classify: " \ --summary_column 'label' \ --max_target_length=4 \ --max_source_length=64 \ --num_train_epochs 20 \ --output_dir ./t5-retrained \ --per_device_train_batch_size=16 \ --per_device_eval_batch_size=16 \ --gradient_accumulation_steps=4 \ --overwrite_output_dir \ --predict_with_generate \ --save_steps=10000 \ --eval_steps=10000 \ --evaluation_strategy steps