#!/bin/bash
##  如果需要指定显卡训练可将 --num_gpus 4 改为 --include="localhost:0,3"
deepspeed --master_port 25678 --num_gpus 4  LLaMA-Factory-main/src/train_bash.py \
    --deepspeed /LLaMA-Factory-main/deepspeed.json \
    --stage sft \
    --do_train \
    --model_name_or_path THUDM/chatglm2-6b \
    --dataset fingpt_sentiment \
    --dataset_dir LLaMA-Factory-main/data \
    --template chatglm3 \
    --finetuning_type lora \
    --lora_target all \
    --output_dir saves/FinGPT/lora_multi_dcu/sft \
    --overwrite_output_dir \
    --cutoff_len 1024 \
    --preprocessing_num_workers 4 \
    --per_device_train_batch_size 4 \
    --per_device_eval_batch_size 4 \
    --gradient_accumulation_steps 1 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --warmup_steps 20 \
    --save_steps 100 \
    --eval_steps 100 \
    --evaluation_strategy steps \
    --load_best_model_at_end \
    --learning_rate 5e-5 \
    --num_train_epochs 3.0 \
    --max_samples 3000 \
    --val_size 0.1 \
    --ddp_timeout 180000000 \
    --plot_loss \
    --fp16