lora_train.sh

export CUDA_VISIBLE_DEVICES=0,1,2,3
deepspeed main.py \
    --deepspeed deepspeed_zero2.json \
    --do_train \
    --train_file AdvertiseGen/train.json \
    --validation_file AdvertiseGen/dev.json \
    --preprocessing_num_workers 10 \
    --prompt_column content \
    --response_column summary \
    --model_name_or_path THUDM/chatglm-6b \
    --output_dir ./output_lora/adgen-chatglm2-6b-lora_version \
    --overwrite_output_dir \
    --max_source_length 64 \
    --max_target_length 64 \
    --per_device_train_batch_size 4 \
    --per_device_eval_batch_size 1 \
    --gradient_accumulation_steps 16 \
    --predict_with_generate \
    --max_steps 10 \
    --logging_steps 10 \
    --save_steps 500 \
    --learning_rate 2e-5 \
    --lora_r 32 \
    --fp16