1  export PATH=/path/to/envs/qwen/bin/:$PATH
 2  cd ./code_arena;
 3  INPUT_PATH=${1}
 4  OUTPUT_PATH=${2}
 5  MODEL_DIR=${3}
 6  TP=${4}
 7  MAX_LEN=${5}
 8  CHAT_TEMPLATE=${6}
 9  CONFIG_FILE=${7}
10  
11  #MAX_LEN=${MAX_LEN:-8192}
12  CHAT_TEMPLATE=${CHAT_TEMPLATE:-"auto"}
13  TP=${TP:-1}
14  INPUT_PATH=${INPUT_PATH:-""}
15  OUTPUT_PATH=${OUTPUT_PATH:-""}
16  MODEL_DIR=${MODEL_DIR:-""}
17  CONFIG_FILE=${CONFIG_FILE:-"./utils/judge_config.yaml"}
18  if [ ! -f ${OUTPUT_PATH} ]; then
19      python infer_vllm.py -model ${MODEL_DIR} -input_path ${INPUT_PATH} -output_path ${OUTPUT_PATH} -tensor_parallel_size ${TP} -model_max_len ${MAX_LEN} -chat_template ${CHAT_TEMPLATE} # 
20  fi
21  EVAL_PATH="${OUTPUT_PATH}.judge"
22  if [ ! -f ${EVAL_PATH} ]; then
23      python judge_models.py -input_path ${OUTPUT_PATH} -output_path ${EVAL_PATH} -workers 64 -judgement_only -setting_file ${CONFIG_FILE} 
24  fi
25  if [ ! -f ${OUTPUT_PATH}.judge.metric ]; then
26      python judge_models.py -input_path ${OUTPUT_PATH} -output_path ${EVAL_PATH} -evaluation_only -setting_file ${CONFIG_FILE}
27  fi