instruction_main.txt

1) create your virtual environment
2) instaill the requirements.txt
   pip install -r requirements_cuda118.txt  (requirements_cuda118_if_the_fist_one_not_working.txt if it is not working)
3) Go to 'dataset_mtr' and read instruction_dataset_mtr.txt'
4) Go to 'models_mtr' and read 'instruction_models_mtr.txt
5) Go to 'finetune' and read 'instruction_finetune.txt'
6) Go to 'eval', then open and run 'Eval_Tabular.ipynb' **

# The finetune result is already located at '/SOME_PATH/transformers_and_chemistry/finetune/evaluations'
# So you can run 'Eval_Tabular.ipynb' ** without running 3)~5)

# The experiments are done under RAM 512, A30 GPU X 4
# You may be required to prepare at least 150GM (or >= 130GB) of RAM if you want to run with four A30 GPUs.
	# Since our enviroment has a large RAM capacity (512 GB) we didn't need to load the dataset with lazy-loading.
	# That's why the MTR training stage requires this much RAM capacity.
	# If you can modify 'datamodule_mtr.py' and 'train_XXX.py' at 'models_mtr' directory regarding data loading to lazay loading 
	# Which is loading data certain amount of rows per each steps from the csv FILES.