增加多进程测试简易方法

benchmark/benchmark.cpp

@@ -5,7 +5,7 @@
 #include "fstream"
 #include <chrono>
 #include "chatglm.h"
-
+#include <unistd.h>
 //static factoryllm fllm;
 //static int modeltype = 0;
 //static char* modelpath = NULL;
@@ -19,6 +19,7 @@ struct BenchmarkConfig {
     int batch = -1; // batch数, -1时使用文件中的行数作为batch
     std::string file; // 输入文件
     std::string output; // 输出文件，如果不设定则输出到屏幕
+    int runloop = 0;
 };
 
 void Usage() {
@@ -51,6 +52,8 @@ void ParseArgs(int argc, char **argv, BenchmarkConfig &config) {
             config.file = sargv[++i];
         } else if (sargv[i] == "-o" || sargv[i] == "--output") {
             config.output = sargv[++i];
+        } else if (sargv[i] == "--loop") {
+            config.runloop = 1;
         } else {
             Usage();
             exit(-1);
@@ -97,31 +100,59 @@ int main(int argc, char **argv) {
         inputs.resize(config.batch);
     }
 
-    std::vector <std::string> outputs;
-    static int tokens = 0;
-    auto st = std::chrono::system_clock::now();
-    chatGlm.ResponseBatch(inputs, outputs, [](int index, std::vector <std::string> &contents) {
-        if (index != -1) {
-            for (int i = 0; i < contents.size(); i++) {
-                tokens += (contents[i].size() > 0);
+    if(config.runloop == 1){
+        while(true){
+            std::vector <std::string> outputs;
+            static int tokens = 0;
+            auto st = std::chrono::system_clock::now();
+            chatGlm.ResponseBatch(inputs, outputs, [](int index, std::vector <std::string> &contents) {
+                if (index != -1) {
+                    for (int i = 0; i < contents.size(); i++) {
+                        tokens += (contents[i].size() > 0);
+                    }
+                }
+            });
+            float spend = GetSpan(st, std::chrono::system_clock::now());
+
+            if (config.output != "") {
+                FILE *fo = fopen(config.output.c_str(), "w");
+                for (int i = 0; i < outputs.size(); i++) {
+                    fprintf(fo, "[ user: \"%s\", model: \"%s\"]\n", inputs[i].c_str(), outputs[i].c_str());
+                }
+                fclose(fo);
             }
+            pid_t pid = getpid();
+//            printf("batch: %d\n", (int)inputs.size());
+            printf("pid %d : output %d tokens\nuse %f s\nspeed = %f tokens / s\n", pid, tokens, spend, tokens / spend);
         }
-    });
-    float spend = GetSpan(st, std::chrono::system_clock::now());
+    }
+    else{
+        std::vector <std::string> outputs;
+        static int tokens = 0;
+        auto st = std::chrono::system_clock::now();
+        chatGlm.ResponseBatch(inputs, outputs, [](int index, std::vector <std::string> &contents) {
+            if (index != -1) {
+                for (int i = 0; i < contents.size(); i++) {
+                    tokens += (contents[i].size() > 0);
+                }
+            }
+        });
+        float spend = GetSpan(st, std::chrono::system_clock::now());
 
-    if (config.output != "") {
-        FILE *fo = fopen(config.output.c_str(), "w");
-        for (int i = 0; i < outputs.size(); i++) {
-            fprintf(fo, "[ user: \"%s\", model: \"%s\"]\n", inputs[i].c_str(), outputs[i].c_str());
-        }
-        fclose(fo);
-    } else {
-        for (int i = 0; i < outputs.size(); i++) {
-            printf("[ user: \"%s\", model: \"%s\"]\n", inputs[i].c_str(), outputs[i].c_str());
+        if (config.output != "") {
+            FILE *fo = fopen(config.output.c_str(), "w");
+            for (int i = 0; i < outputs.size(); i++) {
+                fprintf(fo, "[ user: \"%s\", model: \"%s\"]\n", inputs[i].c_str(), outputs[i].c_str());
+            }
+            fclose(fo);
+        } else {
+            for (int i = 0; i < outputs.size(); i++) {
+                printf("[ user: \"%s\", model: \"%s\"]\n", inputs[i].c_str(), outputs[i].c_str());
+            }
         }
-    }
 
-    printf("batch: %d\n", (int)inputs.size());
-    printf("output %d tokens\nuse %f s\nspeed = %f tokens / s\n", tokens, spend, tokens / spend);
+        printf("batch: %d\n", (int)inputs.size());
+        printf("output %d tokens\nuse %f s\nspeed = %f tokens / s\n", tokens, spend, tokens / spend);
+    }
     return 0;
 }
\ No newline at end of file

benchmark/loop_benchmark.sh

+#!/bin/bash
+
+# 定义要运行的程序命令
+program="./benchmark -p ../../models/chatglm-6b/chatglm_flm/chatglm_6b_test_int8.bin --loop"
+
+# 定义要运行的实例数量
+num_instances=2
+
+# 启动后台进程
+for ((i=1; i<=num_instances; i++)); do
+  $program &
+done
+
+# 压测持续时间（秒）
+test_duration=120
+
+# 等待一段时间以进行压测
+sleep $test_duration
+
+# 杀死所有后台进程
+pkill -f "$program"