# original prompt + qutsion_input + true_option_output
    # data_to_save = [{
    #     "instruction": "Assuming you are a doctor, answer questions based on the patient's symptoms.",
    #     "input": item['question'],
    #     "output": (item['opa'] if item['cop'] == 0 else 
    #                item['opb'] if item['cop'] == 1 else 
    #                item['opc'] if item['cop'] == 2 else 
    #                item['opd'] if item['cop'] == 3 else "No valid answer found")
    # } for item in data]
CUDA_VISIBLE_DEVICES=1 python src/train_bash.py \
    --stage sft \
    --model_name_or_path meta-llama/Llama-2-7b-hf \
    --do_train \
    --dataset alpaca_med_cqa_opn_en \
    --template llama2 \
    --finetuning_type lora \
    --lora_target q_proj,v_proj \
    --output_dir ./FINE/llama2-7b-med_cqa_opn_single \
    --overwrite_cache \
    --per_device_train_batch_size 1 \
    --gradient_accumulation_steps 4 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 1000 \
    --learning_rate 5e-5 \
    --num_train_epochs 3.0 \
    --plot_loss \
    --fp16

# original prompt + qutsion+prompt+options + true_option_output


# original prompt + qutsion_input + true_option_output
    # data_to_save = [{
    #     "instruction": "Assuming you are a doctor, answer questions based on the patient's symptoms.",
    #     "input": item['question']+' \n '+ "select from the following option."+' \n '+ "A. {}, B. {}, C. {}, D. {}".format(item['opa'], item['opb'], item['opc'], item['opd']),
    #     "output": (item['opa'] if item['cop'] == 0 else 
    #                item['opb'] if item['cop'] == 1 else 
    #                item['opc'] if item['cop'] == 2 else 
    #                item['opd'] if item['cop'] == 3 else "No valid answer found")
    # } for item in data]
CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --stage sft \
    --model_name_or_path meta-llama/Llama-2-7b-hf \
    --do_train \
    --dataset alpca_med_cqa_in_modify_ou_T_option_en \
    --template llama2 \
    --finetuning_type lora \
    --lora_target q_proj,v_proj \
    --output_dir ./FINE/llama2-7b-med_cqa_in_modify_ou_T_option_single \
    --overwrite_cache \
    --per_device_train_batch_size 1 \
    --gradient_accumulation_steps 4 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 1000 \
    --learning_rate 5e-5 \
    --num_train_epochs 3.0 \
    --plot_loss \
    --fp16

# modified prompt--qutsion--true output

    # data_to_save = [{
    #     "instruction": "This is real-world medical entrance exam questions, please give the true answer based on the question and selection.",
    #     "input": item['question'],
    #     "output": (item['opa'] if item['cop'] == 0 else 
    #                item['opb'] if item['cop'] == 1 else 
    #                item['opc'] if item['cop'] == 2 else 
    #                item['opd'] if item['cop'] == 3 else "No valid answer found")
    # } for item in data]

CUDA_VISIBLE_DEVICES=2 python src/train_bash.py \
    --stage sft \
    --model_name_or_path meta-llama/Llama-2-7b-hf \
    --do_train \
    --dataset alpca_med_cqa_modi_prompt_q_T_out_en \
    --template llama2 \
    --finetuning_type lora \
    --lora_target q_proj,v_proj \
    --output_dir ./FINE/llama2-7b-med_cqa_modi_prompt_q_T_out_single \
    --overwrite_cache \
    --per_device_train_batch_size 1 \
    --gradient_accumulation_steps 4 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 1000 \
    --learning_rate 5e-5 \
    --num_train_epochs 3.0 \
    --plot_loss \
    --fp16


# modified prompt-- qutsion+prompt+option--true output
    # data_to_save = [{
    #     "instruction": "This is real-world medical entrance exam question, please give the true answer based on the question and selection.",
    #     "input": item['question']+' \n '+ "select from the following option."+' \n '+ "A. {}, B. {}, C. {}, D. {}".format(item['opa'], item['opb'], item['opc'], item['opd']),
    #     "output": (item['opa'] if item['cop'] == 0 else 
    #                item['opb'] if item['cop'] == 1 else 
    #                item['opc'] if item['cop'] == 2 else 
    #                item['opd'] if item['cop'] == 3 else "No valid answer found")
    # } for item in data]

CUDA_VISIBLE_DEVICES=3 python src/train_bash.py \
    --stage sft \
    --model_name_or_path meta-llama/Llama-2-7b-hf \
    --do_train \
    --dataset alpca_med_cqa_modi_prompt_in_modify_ou_T_option_en \
    --template llama2 \
    --finetuning_type lora \
    --lora_target q_proj,v_proj \
    --output_dir ./FINE/llama2-7b-med_cqa_modi_prompt_in_modify_ou_T_option_single \
    --overwrite_cache \
    --per_device_train_batch_size 1 \
    --gradient_accumulation_steps 4 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 1000 \
    --learning_rate 5e-5 \
    --num_train_epochs 3.0 \
    --plot_loss \
    --fp16