TOT_CUDA="0"

CUDAs=(${TOT_CUDA//,/ })

CUDA_NUM=${#CUDAs[@]}

PORT="12345"

DATA_PATH="sample/instruct/data_sample.jsonl" #"../dataset/instruction/guanaco_non_chat_mini_52K-utf8.json"

OUTPUT_PATH="lora-Vicuna-output"

MODEL_PATH="decapoda-research/llama-7b-hf"

lora_remote_checkpoint="Chinese-Vicuna/Chinese-Vicuna-lora-7b-belle-and-guanaco-11600"

lora_checkpoint="./lora-Vicuna/checkpoint-11600"

from_data_beginning=True # False

TEST_SIZE=1

CUDA_VISIBLE_DEVICES=${TOT_CUDA} torchrun --nproc_per_node=$CUDA_NUM --master_port=$PORT finetune.py \

--data_path $DATA_PATH \

--output_path $OUTPUT_PATH \

--model_path $MODEL_PATH \

--eval_steps 200 \

--save_steps 200 \

--test_size $TEST_SIZE \

--resume_from_checkpoint $lora_checkpoint \

--lora_remote_checkpoint $lora_remote_checkpoint \

--ignore_data_skip $from_data_beginning