TOT_CUDA="0,1,2,3"
CUDAs=(${TOT_CUDA//,/ })
CUDA_NUM=${#CUDAs[@]}
PORT="12345"


# ../Chinese-Vicuna/sample/instruct/chat_data.jsonl
#DATA_PATH="sample/instruct/legislation2.json" #"../dataset/instruction/guanaco_non_chat_mini_52K-utf8.json" #"./sample/merge_sample.json"
#DATA_PATH="./sample/instruct/chat_data.jsonl"
DATA_PATH="../Chinese-Vicuna/sample/legislation30k.jsonl"

#DATA_PATH="../Chinese-Vicuna/sample/instructchat_data.jsonl" #working
OUTPUT_PATH="../llama2-30kjudgement-21sept"
MODEL_PATH="../chinese-llama-2-13b"
# lora_checkpoint="../Llama2-Chinese-13b-Chat-LoRA"
from_data_beginning=True
TEST_SIZE=300

# TORCH_DISTRIBUTED_DEBUG=DETAIL 


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

CUDA_VISIBLE_DEVICES=0 python 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