add big tensor monitor config (#592)

Author: Zeref996

test_pipline/V2/monitor_script/run_gpu_bigtensor_full.sh

@@ -0,0 +1,83 @@
+#!/bin/bash
+
+# Script to run engineV2.py
+# Usage: ./run.sh
+
+# 配置参数
+# NUM_GPUS!=0 时，engineV2 不受外部 "CUDA_VISIBLE_DEVICES" 影响
+# FILE_INPUT="tester/api_config/8_big_tensor/big_tensor_1_8.txt"
+FILE_PATTERN="tester/api_config/monitor_config/big_tensor/GPU/monitoring_configs*.txt"
+LOG_DIR="tester/api_config/test_log_gpu_bigtensor_full"
+NUM_GPUS=-1
+NUM_WORKERS_PER_GPU=1
+GPU_IDS="-1"
+# REQUIRED_MEMORY=10
+
+TEST_MODE_ARGS=(
+	--accuracy=True
+	# --paddle_only=True
+    # --paddle_cinn=True
+	# --test_amp=True
+	# --test_cpu=True
+	--use_cached_numpy=True
+)
+
+IN_OUT_ARGS=(
+    # --api_config_file="$FILE_INPUT"
+    --api_config_file_pattern="$FILE_PATTERN"
+    --log_dir="$LOG_DIR"
+)
+
+PARALLEL_ARGS=(
+    --num_gpus="$NUM_GPUS"
+    --num_workers_per_gpu="$NUM_WORKERS_PER_GPU"
+    --gpu_ids="$GPU_IDS"
+    # --required_memory="$REQUIRED_MEMORY"
+)
+
+mkdir -p "$LOG_DIR" || {
+    echo "错误：无法创建日志目录 '$LOG_DIR'"
+    exit 1
+}
+
+# 执行程序
+LOG_FILE="$LOG_DIR/log_$(date +%Y%m%d_%H%M%S).log"
+if [ "$backprocess" -eq 1 ]; then
+    nohup python engineV2.py \
+            "${TEST_MODE_ARGS[@]}" \
+            "${IN_OUT_ARGS[@]}" \
+            "${PARALLEL_ARGS[@]}" \
+            >> "$LOG_FILE" 2>&1 &
+    PYTHON_PID=$!
+
+    sleep 1
+    if ! ps -p "$PYTHON_PID" > /dev/null; then
+        echo "错误：engineV2 启动失败，请检查 $LOG_FILE"
+        exit 1
+    fi
+
+    echo -e "\n\033[32m执行中... 另开终端运行监控:\033[0m"
+    echo -e "1. GPU使用:   watch -n 1 nvidia-smi"
+    echo -e "2. 日志目录:  ls -lh $LOG_DIR"
+    echo -e "3. 详细日志:  tail -f $LOG_FILE"
+    echo -e "4. 终止任务:  kill $PYTHON_PID"
+    echo -e "\n进程已在后台运行，关闭终端不会影响进程执行"
+else
+    python engineV2.py \
+            "${TEST_MODE_ARGS[@]}" \
+            "${IN_OUT_ARGS[@]}" \
+            "${PARALLEL_ARGS[@]}" \
+            2>&1 | tee -a "$LOG_FILE"
+
+    PYTHON_PID=$!
+
+    sleep 1
+    if ! ps -p "$PYTHON_PID" > /dev/null; then
+        echo "错误：engineV2 执行失败，请检查 $LOG_FILE"
+        exit 1
+    fi
+fi
+
+exit 0
+
+# watch -n 1 nvidia-smi --query-compute-apps=pid,process_name,used_memory,gpu_uuid --format=csv