CUDA_VISIBLE_DEVICES=0,1,2 deepspeed --num_gpus=3 src/train_sft.py \
    --deepspeed deepspeed_v2.json \
    --model_name_or_path /home/majiahui/models-LLM/openbuddy-llama-7b-v1.4-fp16/ \
    --do_train \
    --dataset paper_data_v3_prompt \
    --dataset_dir /home/majiahui/project2/LLaMA-Efficient-Tuning/data \
    --finetuning_type freeze \
    --max_source_length 700 \
    --max_target_length 1400 \
    --output_dir path_to_sft_checkpoint_paper_prompt_freeze_deepspeed \
    --overwrite_cache \
    --per_device_train_batch_size 2 \
    --gradient_accumulation_steps 1 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 4000 \
    --learning_rate 1e-5 \
    --num_train_epochs 3.0 \
    --plot_loss \
    --fp16 \
    --checkpoint_dir /home/majiahui/project2/LLaMA-Efficient-Tuning/path_to_sft_checkpoint_paper_prompt_freeze/checkpoint-104000