【TEST】补充math和agieval数据集的冒烟用例

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_agieval/ais_bench_configs/datasets/agieval/accuracy_agieval.py

@@ -0,0 +1,11 @@
+from mmengine import read_base
+
+with read_base():
+    from .agieval_gen_0_shot_chat_prompt import agieval_datasets
+
+# 冒烟：仅取一个子集、小样本
+agieval_datasets = agieval_datasets[:1]
+agieval_datasets[0]['reader_cfg'] = dict(
+    agieval_datasets[0]['reader_cfg'],
+    test_range='[0:10]',
+)

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_agieval/ais_bench_configs/models/vllm_api/accuracy_agieval.py

@@ -0,0 +1,8 @@
+from mmengine import read_base
+
+with read_base():
+    from .vllm_api_general_chat import models
+
+models[0]['model'] = "qwen"
+models[0]['max_out_len'] = 64
+models[0]['batch_size'] = 16

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_agieval/case.yml

@@ -0,0 +1,13 @@
+case_type: benchmark
+case_group:
+  - run_server_accuracy
+  - llm_datasets_main
+  - short
+  - refactor_success_1202
+case_name: accuracy_agieval # 在整个工程的case中必须唯一
+enable: y
+script:
+  start: run.sh
+  end: clean.sh
+timeout: 90
+rank_size: 3

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_agieval/clean.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+CUR_DIR=$(dirname $(readlink -f $0))
+[ -f "${CUR_DIR}/tmplog.txt" ] && rm -f "${CUR_DIR}/tmplog.txt"
+exit 0

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_agieval/run.sh

@@ -0,0 +1,72 @@
+#!/bin/bash
+declare -i ret_ok=0
+declare -i ret_failed=1
+
+CUR_DIR=$(dirname $(readlink -f $0))
+CASE_NAME=$(basename "$CUR_DIR")
+LAST_3_DIRNAME=$(echo  $CUR_DIR | rev | cut -d'/' -f1-3 | rev)
+CASE_OUTPUT_PATH=${PROJECT_OUTPUT_PATH}/${LAST_3_DIRNAME}
+AIS_BENCH_CODE_CONFIGS_DIR=${PROJECT_PATH}/../ais_bench/benchmark/configs
+CONFIG_DATASET_NAME="agieval"
+# 冒烟仅跑一个子集，输出名为该子集 abbr（agieval-gaokao-chinese）
+OUTPUT_DATASET_NAME="agieval-gaokao-chinese"
+CURR_API="vllm-api-general-chat"
+
+if [ ! -d ${CASE_OUTPUT_PATH} ];then
+    mkdir -p ${CASE_OUTPUT_PATH}
+fi
+rm -rf ${CASE_OUTPUT_PATH}/*
+
+echo "Copying config files..."
+cp -r ${CUR_DIR}/ais_bench_configs/* ${AIS_BENCH_CODE_CONFIGS_DIR}/
+
+{
+    echo ""
+    echo "models[0]['host_ip'] = '${AISBENCH_SMOKE_SERVICE_IP}'"
+    echo "models[0]['host_port'] = ${AISBENCH_SMOKE_SERVICE_PORT}"
+    echo "models[0]['path'] = '${AISBENCH_SMOKE_MODEL_PATH}'"
+} >> "${AIS_BENCH_CODE_CONFIGS_DIR}/models/vllm_api/${CASE_NAME}.py"
+
+echo -e "\033[1;32m[1/1]\033[0m Test case - ${CASE_NAME}"
+
+set -o pipefail
+ais_bench --models ${CASE_NAME} --datasets ${CASE_NAME} --work-dir ${CASE_OUTPUT_PATH} 2>&1 | tee ${CUR_DIR}/tmplog.txt
+if [ $? -ne 0 ]
+then
+    echo "Run $CASE_NAME test: Failed"
+    exit $ret_failed
+fi
+echo "Run $CASE_NAME test: Success"
+
+WORK_DIR_INFO=$(cat ${CUR_DIR}/tmplog.txt | grep 'Current exp folder: ')
+TIMESTAMP="${WORK_DIR_INFO##*/}"
+
+CURR_OUTPUT_PATH=${CASE_OUTPUT_PATH}/${TIMESTAMP}
+LOG_EVAL_OUTPUT_PATH=${CURR_OUTPUT_PATH}/logs/eval/${CURR_API}/${OUTPUT_DATASET_NAME}.out
+LOG_INFER_OUTPUT_PATH=${CURR_OUTPUT_PATH}/logs/infer/${CURR_API}/${OUTPUT_DATASET_NAME}.out
+PREDICTIONS_OUTPUT_PATH=${CURR_OUTPUT_PATH}/predictions/${CURR_API}/${OUTPUT_DATASET_NAME}.jsonl
+RESULTS_OUTPUT_PATH=${CURR_OUTPUT_PATH}/results/${CURR_API}/${OUTPUT_DATASET_NAME}.json
+SUMMARY_OUTPUT_PATH=${CURR_OUTPUT_PATH}/summary/summary_${TIMESTAMP}.csv
+
+if [ ! -f "$LOG_EVAL_OUTPUT_PATH" ];then
+    echo "Can't find $LOG_EVAL_OUTPUT_PATH"
+    exit $ret_failed
+fi
+if [ ! -f "$LOG_INFER_OUTPUT_PATH" ];then
+    echo "Can't find $LOG_INFER_OUTPUT_PATH"
+    exit $ret_failed
+fi
+if [ ! -f "$PREDICTIONS_OUTPUT_PATH" ];then
+    echo "Can't find $PREDICTIONS_OUTPUT_PATH"
+    exit $ret_failed
+fi
+if [ ! -f "$RESULTS_OUTPUT_PATH" ];then
+    echo "Can't find $RESULTS_OUTPUT_PATH"
+    exit $ret_failed
+fi
+if [ ! -f "$SUMMARY_OUTPUT_PATH" ];then
+    echo "Can't find $SUMMARY_OUTPUT_PATH"
+    exit $ret_failed
+fi
+
+exit $ret_ok

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_math/ais_bench_configs/datasets/math/accuracy_math.py

@@ -0,0 +1,10 @@
+from mmengine import read_base
+
+with read_base():
+    from .math_prm800k_500_0shot_cot_gen import math_datasets
+
+# 冒烟：小样本
+math_datasets[0]['reader_cfg'] = dict(
+    math_datasets[0]['reader_cfg'],
+    test_range='[0:10]',
+)

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_math/ais_bench_configs/models/vllm_api/accuracy_math.py

@@ -0,0 +1,8 @@
+from mmengine import read_base
+
+with read_base():
+    from .vllm_api_general_chat import models
+
+models[0]['model'] = "qwen"
+models[0]['max_out_len'] = 512
+models[0]['batch_size'] = 8

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_math/case.yml

@@ -0,0 +1,13 @@
+case_type: benchmark
+case_group:
+  - run_server_accuracy
+  - llm_datasets_main
+  - short
+  - refactor_success_1202
+case_name: accuracy_math # 在整个工程的case中必须唯一
+enable: y
+script:
+  start: run.sh
+  end: clean.sh
+timeout: 90
+rank_size: 3

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_math/clean.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+CUR_DIR=$(dirname $(readlink -f $0))
+[ -f "${CUR_DIR}/tmplog.txt" ] && rm -f "${CUR_DIR}/tmplog.txt"
+exit 0

smoke_tests/test-case/run_server_accuracy/llm_datasets_main/accuracy_math/run.sh

@@ -0,0 +1,71 @@
+#!/bin/bash
+declare -i ret_ok=0
+declare -i ret_failed=1
+
+CUR_DIR=$(dirname $(readlink -f $0))
+CASE_NAME=$(basename "$CUR_DIR")
+LAST_3_DIRNAME=$(echo  $CUR_DIR | rev | cut -d'/' -f1-3 | rev)
+CASE_OUTPUT_PATH=${PROJECT_OUTPUT_PATH}/${LAST_3_DIRNAME}
+AIS_BENCH_CODE_CONFIGS_DIR=${PROJECT_PATH}/../ais_bench/benchmark/configs
+CONFIG_DATASET_NAME="math"
+OUTPUT_DATASET_NAME="math_prm800k_500"
+CURR_API="vllm-api-general-chat"
+
+if [ ! -d ${CASE_OUTPUT_PATH} ];then
+    mkdir -p ${CASE_OUTPUT_PATH}
+fi
+rm -rf ${CASE_OUTPUT_PATH}/*
+
+echo "Copying config files..."
+cp -r ${CUR_DIR}/ais_bench_configs/* ${AIS_BENCH_CODE_CONFIGS_DIR}/
+
+{
+    echo ""
+    echo "models[0]['host_ip'] = '${AISBENCH_SMOKE_SERVICE_IP}'"
+    echo "models[0]['host_port'] = ${AISBENCH_SMOKE_SERVICE_PORT}"
+    echo "models[0]['path'] = '${AISBENCH_SMOKE_MODEL_PATH}'"
+} >> "${AIS_BENCH_CODE_CONFIGS_DIR}/models/vllm_api/${CASE_NAME}.py"
+
+echo -e "\033[1;32m[1/1]\033[0m Test case - ${CASE_NAME}"
+
+set -o pipefail
+ais_bench --models ${CASE_NAME} --datasets ${CASE_NAME} --work-dir ${CASE_OUTPUT_PATH} 2>&1 | tee ${CUR_DIR}/tmplog.txt
+if [ $? -ne 0 ]
+then
+    echo "Run $CASE_NAME test: Failed"
+    exit $ret_failed
+fi
+echo "Run $CASE_NAME test: Success"
+
+WORK_DIR_INFO=$(cat ${CUR_DIR}/tmplog.txt | grep 'Current exp folder: ')
+TIMESTAMP="${WORK_DIR_INFO##*/}"
+
+CURR_OUTPUT_PATH=${CASE_OUTPUT_PATH}/${TIMESTAMP}
+LOG_EVAL_OUTPUT_PATH=${CURR_OUTPUT_PATH}/logs/eval/${CURR_API}/${OUTPUT_DATASET_NAME}.out
+LOG_INFER_OUTPUT_PATH=${CURR_OUTPUT_PATH}/logs/infer/${CURR_API}/${OUTPUT_DATASET_NAME}.out
+PREDICTIONS_OUTPUT_PATH=${CURR_OUTPUT_PATH}/predictions/${CURR_API}/${OUTPUT_DATASET_NAME}.jsonl
+RESULTS_OUTPUT_PATH=${CURR_OUTPUT_PATH}/results/${CURR_API}/${OUTPUT_DATASET_NAME}.json
+SUMMARY_OUTPUT_PATH=${CURR_OUTPUT_PATH}/summary/summary_${TIMESTAMP}.csv
+
+if [ ! -f "$LOG_EVAL_OUTPUT_PATH" ];then
+    echo "Can't find $LOG_EVAL_OUTPUT_PATH"
+    exit $ret_failed
+fi
+if [ ! -f "$LOG_INFER_OUTPUT_PATH" ];then
+    echo "Can't find $LOG_INFER_OUTPUT_PATH"
+    exit $ret_failed
+fi
+if [ ! -f "$PREDICTIONS_OUTPUT_PATH" ];then
+    echo "Can't find $PREDICTIONS_OUTPUT_PATH"
+    exit $ret_failed
+fi
+if [ ! -f "$RESULTS_OUTPUT_PATH" ];then
+    echo "Can't find $RESULTS_OUTPUT_PATH"
+    exit $ret_failed
+fi
+if [ ! -f "$SUMMARY_OUTPUT_PATH" ];then
+    echo "Can't find $SUMMARY_OUTPUT_PATH"
+    exit $ret_failed
+fi
+
+exit $ret_ok