deepspeed --num_gpus=8 src/train_bash.py \
    --stage sft \
    --model_name_or_path ../baichuan-13b-chat \
    --do_train \
    --template baichuan \
    --dataset alpaca_gpt4_en,alpaca_gpt4_zh,self_cognition,oaast_sft,sharegpt_zh,lima \
    --finetuning_type full \
    --output_dir output \
    --per_device_train_batch_size 1 \
    --per_device_eval_batch_size 1 \
    --gradient_accumulation_steps 2 \
    --preprocessing_num_workers 16 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 100 \
    --eval_steps 100 \
    --learning_rate 5e-5 \
    --max_grad_norm 0.5 \
    --num_train_epochs 2.0 \
    --val_size 0.01 \
    --evaluation_strategy steps \
    --weight_decay 0. \
    --warmup_ratio 0.03 \
    --max_source_length 2000 \
    --max_target_length 1200 \
    --plot_loss \
    --fp16 \
    --deepspeed deepspeed.json