diff --git a/benchmark/benchmark.go b/benchmark/benchmark.go index 62a67a9..e7800ea 100644 --- a/benchmark/benchmark.go +++ b/benchmark/benchmark.go @@ -38,6 +38,8 @@ type Choice struct { Message Message `json:"message"` Index int `json:"index"` FinishReason string `json:"finish_reason"` + // reasoning_content: 思维链内容 + ReasoningContent string `json:"reasoning_content"` } // Usage 表示API响应中的使用情况 @@ -69,6 +71,8 @@ type Result struct { ResponseTime time.Duration // 响应时间 FirstTokenTime time.Duration // 首个令牌生成时间(TTFT) Response string // 响应内容 + // 推理模型的思维链内容 + ReasoningContent string Error error // 错误信息 } @@ -107,7 +111,7 @@ func NewBenchmark(cfg *config.Config) (*Benchmark, error) { return &Benchmark{ config: cfg, generator: gen, - httpClient: &http.Client{}, + httpClient: &http.Client{Timeout: time.Duration(cfg.Timeout) * time.Second}, rand: rand.New(rand.NewSource(time.Now().UnixNano())), }, nil } @@ -145,20 +149,23 @@ func (b *Benchmark) executeRequest(concurrencyStep int) Result { promptType := b.generator.SelectPromptType() result.PromptType = promptType - // 生成提示词 + // 生成并统计提示词 prompt, err := b.generator.GeneratePrompt(promptType) if err != nil { result.Error = fmt.Errorf("生成提示词失败: %w", err) return result } + promptTokens := b.generator.CountTokens(prompt) + result.PromptTokenCount = promptTokens + result.PromptTokens = promptTokens - // 创建请求 + // 构造请求对象 reqObj := Request{ Model: b.config.API.Model, Messages: []Message{{Role: "user", Content: prompt}}, - MaxTokens: 2048, - Temperature: 0.7, - Stream: true, // 启用流式响应 + MaxTokens: b.config.API.MaxTokens, + Temperature: b.config.API.Temperature, + Stream: true, } reqBody, err := json.Marshal(reqObj) @@ -168,11 +175,9 @@ func (b *Benchmark) executeRequest(concurrencyStep int) Result { } requestID := fmt.Sprintf("request-%s", result.Timestamp.Format("20060102-150405.000")) - requestLog := fmt.Sprintf("===== 请求 ID: %s =====\n时间: %s\n并发级别: %d\n提示词类型: %s\n端点: %s\n请求体:\n%s\n", - requestID, result.Timestamp.Format(time.RFC3339), concurrencyStep, promptType, b.config.API.Endpoint, string(reqBody)) - logger.LogToFile(requestLog) + logger.LogToFile(fmt.Sprintf("===== 请求 ID: %s =====\n并发: %d, 类型: %s, 端点: %s\n请求体:\n%s\n", requestID, concurrencyStep, promptType, b.config.API.Endpoint, string(reqBody))) - // 构造 http.Request + // 构造 HTTP 请求并绑定超时上下文 httpReq, err := http.NewRequest("POST", b.config.API.Endpoint, bytes.NewReader(reqBody)) if err != nil { result.Error = fmt.Errorf("创建HTTP请求失败: %w", err) @@ -182,26 +187,36 @@ func (b *Benchmark) executeRequest(concurrencyStep int) Result { httpReq.Header.Set("Authorization", "Bearer "+b.config.API.APIKey) httpReq.Header.Set("Accept", "text/event-stream") + ctxReq, cancel := context.WithTimeout(httpReq.Context(), time.Duration(b.config.Timeout)*time.Second) + defer cancel() + httpReq = httpReq.WithContext(ctxReq) + startTime := time.Now() resp, err := b.httpClient.Do(httpReq) if err != nil { result.Error = fmt.Errorf("发送请求失败: %w", err) - logger.Error("请求失败: %v", err) + logger.Error("请求错误: %v", err) return result } defer resp.Body.Close() + // 检查状态码 + if resp.StatusCode != http.StatusOK { + result.Error = fmt.Errorf("HTTP状态码错误: %d", resp.StatusCode) + logger.Warn("HTTP状态码非200: %d", resp.StatusCode) + return result + } + + // 记录响应时间 responseTime := time.Since(startTime) result.ResponseTime = responseTime - // 记录响应日志 - respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 4096)) // 只记录前4K - responseLog := fmt.Sprintf("===== 响应 ID: %s =====\n时间: %s\n响应时间: %v\n状态码: %d\n响应头:\n%s\n响应体(前4K):\n%s\n", - requestID, time.Now().Format(time.RFC3339), responseTime, resp.StatusCode, resp.Header, string(respBody)) - logger.LogToFile(responseLog) - - // 重置 resp.Body 以便流式处理 - resp.Body = io.NopCloser(io.MultiReader(bytes.NewReader(respBody), resp.Body)) + // 非流式记录前4K + if !reqObj.Stream { + head, _ := io.ReadAll(io.LimitReader(resp.Body, 4096)) + logger.LogToFile(fmt.Sprintf("===== 响应头部 ID: %s =====\n状态码: %d\n前4K响应体:\n%s\n", requestID, resp.StatusCode, string(head))) + resp.Body = io.NopCloser(io.MultiReader(bytes.NewReader(head), resp.Body)) + } if reqObj.Stream { return b.handleStreamResponse(resp, result, startTime) @@ -230,12 +245,13 @@ func (b *Benchmark) runConcurrencyStep(concurrencyStep int) ([]Result, error) { case <-ctx.Done(): return default: - result := b.executeRequest(concurrencyStep) + res := b.executeRequest(concurrencyStep) resultsMutex.Lock() - results = append(results, result) + results = append(results, res) resultsMutex.Unlock() - // 简单限速,可根据需要调整 - time.Sleep(10 * time.Millisecond) + // 按泊松分布间隔 + interval := b.rand.ExpFloat64() / b.config.PoissonLambda + time.Sleep(time.Duration(interval * float64(time.Second))) } } }(i) @@ -245,177 +261,105 @@ func (b *Benchmark) runConcurrencyStep(concurrencyStep int) ([]Result, error) { return results, nil } -// handleStreamResponse 处理流式响应(net/http 实现) +// handleStreamResponse 处理流式响应 func (b *Benchmark) handleStreamResponse(resp *http.Response, result Result, startTime time.Time) Result { - responseContent := "" + var builder strings.Builder + // 聚合思维链 + var reasoningBuilder strings.Builder tokenCount := 0 firstTokenReceived := false var firstTokenTime time.Duration - - // 记录响应头接收时间 - 这是我们开始计算TTFT的时间点 + + reader := bufio.NewReaderSize(resp.Body, 4096) responseHeaderTime := time.Now() logger.Debug("接收到响应头,网络请求耗时: %v", responseHeaderTime.Sub(startTime)) - - // 创建一个带缓冲的读取器,这样我们可以更快地读取数据 - reader := bufio.NewReaderSize(resp.Body, 4096) - - requestID := fmt.Sprintf("request-%s", result.Timestamp.Format("20060102-150405.000")) - streamLogHeader := fmt.Sprintf("===== 流式响应开始 ID: %s =====\n时间: %s\n响应头: %v\n", - requestID, responseHeaderTime.Format(time.RFC3339), resp.Header) - logger.LogToFile(streamLogHeader) - - // 使用更高效的方式处理流式响应 + chunkIdx := 0 - buffer := make([]byte, 1024) - var line []byte - for { - n, err := reader.Read(buffer) + line, err := reader.ReadString('\n') if err != nil { if err != io.EOF { - logger.Warn("读取响应流出错: %v", err) + logger.Warn("读取流式响应出错: %v", err) } break } - - // 处理读取到的数据 - data := buffer[:n] - lines := bytes.Split(data, []byte("\n")) - - for _, currentLine := range lines { - if len(currentLine) == 0 { - continue - } - - // 如果行不完整,先缓存起来 - if !bytes.HasSuffix(currentLine, []byte("\n")) { - line = append(line, currentLine...) - continue - } - - // 处理完整的行 - line = append(line, currentLine...) - - // 检查是否为数据行 - if !bytes.HasPrefix(line, []byte("data: ")) { - line = nil - continue - } - - // 移除 "data: " 前缀 - data := line[6:] - line = nil - - // 检查是否为结束标记 - if bytes.Equal(data, []byte("[DONE]")) { - logger.Debug("收到流式响应结束标记") - break - } - - // 解析JSON - var chunk map[string]interface{} - if err := json.Unmarshal(data, &chunk); err != nil { - logger.Warn("解析JSON数据块失败: %v, 数据: %s", err, string(data)) - continue - } - - // 提取内容 - choices, ok := chunk["choices"].([]interface{}) - if !ok || len(choices) == 0 { - continue - } - - choice, ok := choices[0].(map[string]interface{}) - if !ok { - continue - } - - delta, ok := choice["delta"].(map[string]interface{}) - if !ok { - continue - } - - content, ok := delta["content"].(string) - if !ok { - continue - } - - // 记录当前时间 - now := time.Now() - - responseContent += content - tokenCount += len(strings.Split(content, " ")) - chunkIdx++ - chunkLog := fmt.Sprintf("[Chunk %d] %s 内容: %s\n", chunkIdx, now.Format(time.RFC3339Nano), content) - logger.LogToFile(chunkLog) - - // 记录首个令牌时间 - 从响应头接收时间开始计算 - if !firstTokenReceived && len(content) > 0 { + if !strings.HasPrefix(line, "data: ") { + continue + } + data := strings.TrimPrefix(line, "data: ") + data = strings.TrimSuffix(data, "\n") + if data == "[DONE]" { + logger.Debug("收到流式响应结束标记") + break + } + var chunk map[string]interface{} + if err := json.Unmarshal([]byte(data), &chunk); err != nil { + logger.Warn("JSON解析失败: %v", err) + continue + } + choices, ok := chunk["choices"].([]interface{}) + if !ok || len(choices) == 0 { + continue + } + choiceIface := choices[0] + choiceMap, ok := choiceIface.(map[string]interface{}) + if !ok { + continue + } + deltaIface, ok := choiceMap["delta"] + deltaMap, ok := deltaIface.(map[string]interface{}) + if !ok { + continue + } + contentVal, ok := deltaMap["content"] + content, ok := contentVal.(string) + if !ok || content == "" { + continue + } + now := time.Now() + builder.WriteString(content) + tokenCount += len(strings.Split(content, " ")) + chunkIdx++ + logger.LogToFile(fmt.Sprintf("[Chunk %d] %s 内容: %s", chunkIdx, now.Format(time.RFC3339Nano), content)) + // 初次令牌检测:推理模型优先考虑思维链内容 + rcVal, rcOk := deltaMap["reasoning_content"] + rcStr, _ := rcVal.(string) + if !firstTokenReceived { + if rcOk && rcStr != "" { firstTokenTime = now.Sub(responseHeaderTime) firstTokenReceived = true - logger.Debug("收到首个令牌,TTFT: %v (从接收响应头开始计算)", firstTokenTime) - - // 记录更详细的TTFT信息 - ttftLog := fmt.Sprintf("===== TTFT详情 ID: %s =====\n请求开始时间: %s\n响应头接收时间: %s\n首个令牌接收时间: %s\n总请求耗时: %v\n网络请求耗时: %v\nTTFT(从响应头计算): %v\n", - requestID, - startTime.Format(time.RFC3339Nano), - responseHeaderTime.Format(time.RFC3339Nano), - now.Format(time.RFC3339Nano), - now.Sub(startTime), - responseHeaderTime.Sub(startTime), - firstTokenTime) - logger.LogToFile(ttftLog) + logger.Debug("收到首个思维链令牌,TTFT: %v", firstTokenTime) + } else if content != "" { + firstTokenTime = now.Sub(responseHeaderTime) + firstTokenReceived = true + logger.Debug("收到首个内容令牌,TTFT: %v", firstTokenTime) } } - } - - streamLogFooter := fmt.Sprintf("===== 流式响应结束 ID: %s =====\n累计响应内容: %s\n", requestID, responseContent) - logger.LogToFile(streamLogFooter) - - // 计算总响应时间 - result.ResponseTime = time.Since(startTime) - - // 设置TTFT - if firstTokenReceived { - // 使用更精确的TTFT计算方式 - // 实际测试中,我们发现首个令牌的生成时间应该非常短 - // 如果测量值超过1秒,可能是由于网络延迟或其他因素导致的 - // 因此,我们使用一个更合理的估计值 - if firstTokenTime > 1*time.Second { - // 使用更合理的估计值,基于实际测试结果 - estimatedTTFT := 200 * time.Millisecond - logger.Debug("测量的TTFT(%v)可能不准确,使用估计值: %v", firstTokenTime, estimatedTTFT) - result.FirstTokenTime = estimatedTTFT - } else { - result.FirstTokenTime = firstTokenTime + // 聚合思维链字段 + if rcOk && rcStr != "" { + reasoningBuilder.WriteString(rcStr) } - } else { - // 如果没有检测到首个令牌,使用默认估计值 - estimatedTTFT := 500 * time.Millisecond - logger.Warn("未检测到首个令牌,使用估计值作为TTFT: %v", estimatedTTFT) - result.FirstTokenTime = estimatedTTFT } - - // 设置结果 + responseContent := builder.String() + result.Response = responseContent - result.PromptTokenCount = len(result.PromptType) / 2 // 简单估计 - result.PromptTokens = len(result.PromptType) / 2 // 简单估计 + // CompletionTokens 和总Token result.CompletionTokens = tokenCount result.TotalTokens = result.PromptTokens + result.CompletionTokens - - // 记录处理后的响应 - processedLog := fmt.Sprintf("===== 处理后的响应 ID: request-%s =====\n响应时间: %v\nTTFT: %v\n提示词Token: %d\n完成Token: %d\n总Token: %d\n响应内容:\n%s\n", - result.Timestamp.Format("20060102-150405.000"), result.ResponseTime, result.FirstTokenTime, - result.PromptTokens, result.CompletionTokens, result.TotalTokens, result.Response) - logger.LogToFile(processedLog) - - logger.Debug("流式响应处理完成,总响应时间: %v, TTFT: %v, 提示词Token: %d, 完成Token: %d, 总Token: %d", - result.ResponseTime, result.FirstTokenTime, result.PromptTokens, result.CompletionTokens, result.TotalTokens) - + // 设置思维链内容 + result.ReasoningContent = reasoningBuilder.String() + // 设置响应时间和首个令牌时间 (TTFT) + result.ResponseTime = time.Since(startTime) + if firstTokenReceived { + result.FirstTokenTime = firstTokenTime + } else { + // 未检测到首个令牌,使用响应时间作为 TTFT + result.FirstTokenTime = result.ResponseTime + } return result } -// handleNormalResponse 处理普通响应(net/http 实现) +// handleNormalResponse 处理普通响应 func (b *Benchmark) handleNormalResponse(resp *http.Response, result Result) Result { var response Response if err := json.NewDecoder(resp.Body).Decode(&response); err != nil { @@ -423,11 +367,23 @@ func (b *Benchmark) handleNormalResponse(resp *http.Response, result Result) Res logger.Error("解析响应失败: %v", err) return result } - result.PromptTokens = response.Usage.PromptTokens - result.CompletionTokens = response.Usage.CompletionTokens - result.TotalTokens = response.Usage.TotalTokens + // 先获取响应内容 if len(response.Choices) > 0 { - result.Response = response.Choices[0].Message.Content + choice := response.Choices[0] + result.Response = choice.Message.Content + // 推理模型思维链 + result.ReasoningContent = choice.ReasoningContent + } + // 根据 Usage 判断是否兼容推理模型 + if response.Usage.TotalTokens > 0 { + result.PromptTokens = response.Usage.PromptTokens + result.CompletionTokens = response.Usage.CompletionTokens + result.TotalTokens = response.Usage.TotalTokens + } else { + // 推理模型未返回 Usage,使用本地统计 + result.PromptTokens = result.PromptTokenCount + result.CompletionTokens = b.generator.CountTokens(result.Response) + result.TotalTokens = result.PromptTokens + result.CompletionTokens } result.FirstTokenTime = result.ResponseTime logger.Debug("非流式响应,使用响应时间作为TTFT: %v", result.FirstTokenTime) diff --git a/config/config.go b/config/config.go index e35896e..c20855c 100644 --- a/config/config.go +++ b/config/config.go @@ -18,9 +18,11 @@ type Config struct { // APIConfig 表示API相关配置 type APIConfig struct { - Endpoint string `mapstructure:"endpoint"` - APIKey string `mapstructure:"api_key"` - Model string `mapstructure:"model"` + Endpoint string `mapstructure:"endpoint"` + APIKey string `mapstructure:"api_key"` + Model string `mapstructure:"model"` + MaxTokens int `mapstructure:"max_tokens"` + Temperature float64 `mapstructure:"temperature"` } // PromptTemplatesConfig 表示提示词模板配置 @@ -70,6 +72,14 @@ func LoadConfig(path string) (*Config, error) { return nil, err } + // 设置默认值 + if config.API.MaxTokens == 0 { + config.API.MaxTokens = 2048 + } + if config.API.Temperature == 0 { + config.API.Temperature = 0.7 + } + return &config, nil } diff --git a/llm-benchmark b/llm-benchmark index 205dd30..a337573 100755 Binary files a/llm-benchmark and b/llm-benchmark differ diff --git a/prompt/generator.go b/prompt/generator.go index 72b8a6f..43b5261 100644 --- a/prompt/generator.go +++ b/prompt/generator.go @@ -325,3 +325,9 @@ func (g *Generator) SelectPromptType() string { // 默认返回第一个类型 return g.config.Requests[0].Type } + +// CountTokens 返回字符串的 token 数 +func (g *Generator) CountTokens(s string) int { + tokens := g.tokenizer.Encode(s, nil, nil) + return len(tokens) +} diff --git a/report/report.go b/report/report.go index 5e5eb86..2d0f8e8 100644 --- a/report/report.go +++ b/report/report.go @@ -87,7 +87,7 @@ func GenerateReport(results *benchmark.BenchmarkResults, cfg *config.Config, out // 生成CSV数据 logger.Info("生成CSV数据...") - if err := generateCSVData(results, outputPath); err != nil { + if err := generateCSVData(results, metrics, outputPath); err != nil { return fmt.Errorf("生成CSV数据失败: %w", err) } @@ -343,13 +343,16 @@ func createOverviewChart(results *benchmark.BenchmarkResults, metrics *Metrics, // 添加数据 chart.SetXAxis([]string{ - "总请求数", - "成功请求数", - "失败请求数", - "平均响应时间(ms)", - "P90响应时间(ms)", - "P95响应时间(ms)", + "总请求数", + "成功请求数", + "失败请求数", + "超时比率(%)", + "平均响应时间(ms)", + "P90响应时间(ms)", + "P95响应时间(ms)", "P99响应时间(ms)", + "最小响应时间(ms)", + "最大响应时间(ms)", "最小TTFT(ms)", "最大TTFT(ms)", "P90 TTFT(ms)", @@ -357,16 +360,22 @@ func createOverviewChart(results *benchmark.BenchmarkResults, metrics *Metrics, "P99 TTFT(ms)", "平均QPS", "最大QPS", + "平均Token生成速率(tokens/s)", + "最大Token生成速率(tokens/s)", + "最大有效并发用户数", }) chart.AddSeries("指标", []opts.BarData{ {Value: metrics.TotalRequests}, {Value: metrics.SuccessfulRequests}, {Value: metrics.FailedRequests}, + {Value: metrics.TimeoutRatio * 100}, {Value: metrics.AvgResponseTime.Milliseconds()}, {Value: metrics.P90ResponseTime.Milliseconds()}, {Value: metrics.P95ResponseTime.Milliseconds()}, {Value: metrics.P99ResponseTime.Milliseconds()}, + {Value: metrics.MinResponseTime.Milliseconds()}, + {Value: metrics.MaxResponseTime.Milliseconds()}, {Value: metrics.MinTTFT.Milliseconds()}, {Value: metrics.MaxTTFT.Milliseconds()}, {Value: metrics.P90TTFT.Milliseconds()}, @@ -374,6 +383,9 @@ func createOverviewChart(results *benchmark.BenchmarkResults, metrics *Metrics, {Value: metrics.P99TTFT.Milliseconds()}, {Value: metrics.AvgQPS}, {Value: metrics.MaxQPS}, + {Value: metrics.AvgTokenRate}, + {Value: metrics.MaxTokenRate}, + {Value: metrics.MaxConcurrency}, }) return chart @@ -721,7 +733,7 @@ func createTTFTOverTimeChart(results *benchmark.BenchmarkResults) *charts.Line { } // generateCSVData 生成CSV数据 -func generateCSVData(results *benchmark.BenchmarkResults, outputPath string) error { +func generateCSVData(results *benchmark.BenchmarkResults, metrics *Metrics, outputPath string) error { // 创建CSV文件 csvPath := filepath.Join(outputPath, "results.csv") logger.Debug("创建CSV文件: %s", csvPath) @@ -733,7 +745,7 @@ func generateCSVData(results *benchmark.BenchmarkResults, outputPath string) err // 写入CSV头 logger.Debug("写入CSV头...") - _, err = f.WriteString("Timestamp,PromptType,PromptTokens,CompletionTokens,ResponseTime,Concurrency,FirstTokenTime,Error\n") + _, err = f.WriteString("Timestamp,PromptType,PromptTokens,CompletionTokens,TotalTokens,ResponseTime(ms),FirstTokenTime(ms),Concurrency,Error,ReasoningContent\n") if err != nil { return err } @@ -754,15 +766,17 @@ func generateCSVData(results *benchmark.BenchmarkResults, outputPath string) err } // 写入一行数据 - line := fmt.Sprintf("%s,%s,%d,%d,%d,%d,%d,%s\n", + line := fmt.Sprintf("%s,%s,%d,%d,%d,%d,%d,%d,%s,%q\n", result.Timestamp.Format(time.RFC3339), result.PromptType, result.PromptTokens, result.CompletionTokens, + result.TotalTokens, result.ResponseTime.Milliseconds(), - result.Concurrency, result.FirstTokenTime.Milliseconds(), + result.Concurrency, errorMsg, + result.ReasoningContent, ) _, err = f.WriteString(line) diff --git a/reports/report_2025-04-22_15-04-25/api_logs.txt b/reports/report_2025-04-22_15-04-25/api_logs.txt new file mode 100644 index 0000000..4f54391 --- /dev/null +++ b/reports/report_2025-04-22_15-04-25/api_logs.txt @@ -0,0 +1,55 @@ +[2025-04-22 15:04:25.267] [信息] 日志文件已设置: reports/report_2025-04-22_15-04-25/api_logs.txt +[2025-04-22 15:04:25.267] [信息] API请求响应日志将记录到: reports/report_2025-04-22_15-04-25/api_logs.txt +[2025-04-22 15:04:25.267] [信息] 创建报告输出目录: reports/report_2025-04-22_15-04-25 +[2025-04-22 15:04:25.267] [调试] 共有 3 条测试结果记录 +[2025-04-22 15:04:25.267] [信息] 计算性能指标... +[2025-04-22 15:04:25.267] [调试] 计算指标 - 总请求数: 3 +[2025-04-22 15:04:25.267] [调试] 计算指标 - 成功请求: 3, 失败请求: 0 +[2025-04-22 15:04:25.267] [调试] 计算指标 - 超时比率: 0.00% +[2025-04-22 15:04:25.267] [调试] 计算指标 - 响应时间(ms) - 平均: 166.00, P90: 185.00, P95: 185.00, P99: 185.00, 最小: 128.00, 最大: 214.00 +[2025-04-22 15:04:25.267] [调试] 计算指标 - TTFT(ms) - 最小: 0.00, 最大: 0.00, P90: 0.00, P95: 0.00, P99: 0.00 +[2025-04-22 15:04:25.267] [调试] 计算指标 - 平均QPS: 0.04 (总持续时间: 69.39秒) +[2025-04-22 15:04:25.267] [调试] 计算各并发步骤的QPS... +[2025-04-22 15:04:25.267] [调试] 并发步骤 1: 持续时间过短 +[2025-04-22 15:04:25.267] [调试] 并发步骤 2: QPS = 214270.41 (成功请求: 2, 持续时间: 0.00秒) +[2025-04-22 15:04:25.267] [调试] 更新最大QPS: 214270.41 (并发步骤: 2) +[2025-04-22 15:04:25.267] [调试] 更新最大有效并发用户数: 2 +[2025-04-22 15:04:25.267] [调试] 计算指标 - 平均Token生成速率: 22.12 tokens/s (总Token: 1535) +[2025-04-22 15:04:25.267] [调试] 计算各并发步骤的Token生成速率... +[2025-04-22 15:04:25.267] [调试] 并发步骤 2: Token生成速率 = 149453610.46 tokens/s (总Token: 1395, 持续时间: 0.00秒) +[2025-04-22 15:04:25.267] [调试] 更新最大Token生成速率: 149453610.46 tokens/s (并发步骤: 2) +[2025-04-22 15:04:25.267] [信息] 性能指标摘要: +[2025-04-22 15:04:25.267] [信息] 总请求数: 3 (成功: 3, 失败: 0) +[2025-04-22 15:04:25.267] [信息] 超时比率: 0.00% +[2025-04-22 15:04:25.267] [信息] 平均响应时间: 166ms +[2025-04-22 15:04:25.267] [信息] P90响应时间: 185ms +[2025-04-22 15:04:25.267] [信息] P95响应时间: 185ms +[2025-04-22 15:04:25.267] [信息] P99响应时间: 185ms +[2025-04-22 15:04:25.267] [信息] 平均QPS: 0.04 +[2025-04-22 15:04:25.268] [信息] 最大QPS: 214270.41 +[2025-04-22 15:04:25.268] [信息] 平均Token生成速率: 22.12 tokens/s +[2025-04-22 15:04:25.268] [信息] 最大Token生成速率: 149453610.46 tokens/s +[2025-04-22 15:04:25.268] [信息] 最大有效并发用户数: 2 +[2025-04-22 15:04:25.268] [信息] 最小首个令牌生成时间: 0s +[2025-04-22 15:04:25.268] [信息] 最大首个令牌生成时间: 0s +[2025-04-22 15:04:25.268] [信息] P90首个令牌生成时间: 0s +[2025-04-22 15:04:25.268] [信息] P95首个令牌生成时间: 0s +[2025-04-22 15:04:25.268] [信息] P99首个令牌生成时间: 0s +[2025-04-22 15:04:25.268] [信息] 生成HTML报告... +[2025-04-22 15:04:25.268] [调试] 创建HTML报告页面... +[2025-04-22 15:04:25.268] [调试] 添加概述图表... +[2025-04-22 15:04:25.268] [调试] 添加响应时间分布图... +[2025-04-22 15:04:25.268] [调试] 添加QPS随时间变化图... +[2025-04-22 15:04:25.268] [调试] 添加Token生成速率随时间变化图... +[2025-04-22 15:04:25.268] [调试] 添加并发与响应时间关系图... +[2025-04-22 15:04:25.268] [调试] 添加TTFT随时间变化图... +[2025-04-22 15:04:25.268] [调试] 保存HTML报告: reports/report_2025-04-22_15-04-25/report.html +[2025-04-22 15:04:25.272] [信息] 生成CSV数据... +[2025-04-22 15:04:25.272] [调试] 创建CSV文件: reports/report_2025-04-22_15-04-25/results.csv +[2025-04-22 15:04:25.272] [调试] 写入CSV头... +[2025-04-22 15:04:25.272] [调试] 写入CSV数据,共3条记录... +[2025-04-22 15:04:25.272] [调试] CSV数据写入完成 +[2025-04-22 15:04:25.272] [信息] 报告生成完成 +[2025-04-22 15:04:25.272] [信息] 基准测试完成!报告已生成: reports/report_2025-04-22_15-04-25 +[2025-04-22 15:04:25.272] [信息] HTML报告: reports/report_2025-04-22_15-04-25/report.html +[2025-04-22 15:04:25.272] [信息] CSV数据: reports/report_2025-04-22_15-04-25/results.csv diff --git a/reports/report_2025-04-22_15-04-25/report.html b/reports/report_2025-04-22_15-04-25/report.html new file mode 100644 index 0000000..e829597 --- /dev/null +++ b/reports/report_2025-04-22_15-04-25/report.html @@ -0,0 +1,121 @@ + + + + + + LLM API 基准测试报告 + + + + + + + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + + + + + + diff --git a/reports/report_2025-04-22_15-04-25/results.csv b/reports/report_2025-04-22_15-04-25/results.csv new file mode 100644 index 0000000..2033909 --- /dev/null +++ b/reports/report_2025-04-22_15-04-25/results.csv @@ -0,0 +1,4 @@ +Timestamp,PromptType,PromptTokens,CompletionTokens,ResponseTime,Concurrency,FirstTokenTime,Error +2025-04-22T15:03:15+08:00,short,52,140,156,1,0, +2025-04-22T15:03:34+08:00,short,48,363,128,2,0, +2025-04-22T15:03:34+08:00,long,960,1032,214,2,0, diff --git a/reports/report_2025-04-22_15-17-42/api_logs.txt b/reports/report_2025-04-22_15-17-42/api_logs.txt new file mode 100644 index 0000000..6aca2a3 --- /dev/null +++ b/reports/report_2025-04-22_15-17-42/api_logs.txt @@ -0,0 +1,55 @@ +[2025-04-22 15:17:42.168] [信息] 日志文件已设置: reports/report_2025-04-22_15-17-42/api_logs.txt +[2025-04-22 15:17:42.168] [信息] API请求响应日志将记录到: reports/report_2025-04-22_15-17-42/api_logs.txt +[2025-04-22 15:17:42.168] [信息] 创建报告输出目录: reports/report_2025-04-22_15-17-42 +[2025-04-22 15:17:42.168] [调试] 共有 3 条测试结果记录 +[2025-04-22 15:17:42.168] [信息] 计算性能指标... +[2025-04-22 15:17:42.168] [调试] 计算指标 - 总请求数: 3 +[2025-04-22 15:17:42.169] [调试] 计算指标 - 成功请求: 3, 失败请求: 0 +[2025-04-22 15:17:42.169] [调试] 计算指标 - 超时比率: 0.00% +[2025-04-22 15:17:42.169] [调试] 计算指标 - 响应时间(ms) - 平均: 22476.33, P90: 24723.00, P95: 24723.00, P99: 24723.00, 最小: 17983.00, 最大: 24791.00 +[2025-04-22 15:17:42.169] [调试] 计算指标 - TTFT(ms) - 最小: 9267.00, 最大: 11273.00, P90: 11088.00, P95: 11088.00, P99: 11088.00 +[2025-04-22 15:17:42.169] [调试] 计算指标 - 平均QPS: 0.06 (总持续时间: 50.92秒) +[2025-04-22 15:17:42.169] [调试] 计算各并发步骤的QPS... +[2025-04-22 15:17:42.169] [调试] 并发步骤 1: 持续时间过短 +[2025-04-22 15:17:42.169] [调试] 并发步骤 2: QPS = 205128.21 (成功请求: 2, 持续时间: 0.00秒) +[2025-04-22 15:17:42.169] [调试] 更新最大QPS: 205128.21 (并发步骤: 2) +[2025-04-22 15:17:42.169] [调试] 更新最大有效并发用户数: 2 +[2025-04-22 15:17:42.169] [调试] 计算指标 - 平均Token生成速率: 19.25 tokens/s (总Token: 980) +[2025-04-22 15:17:42.169] [调试] 计算各并发步骤的Token生成速率... +[2025-04-22 15:17:42.169] [调试] 并发步骤 2: Token生成速率 = 68307692.31 tokens/s (总Token: 666, 持续时间: 0.00秒) +[2025-04-22 15:17:42.169] [调试] 更新最大Token生成速率: 68307692.31 tokens/s (并发步骤: 2) +[2025-04-22 15:17:42.169] [信息] 性能指标摘要: +[2025-04-22 15:17:42.169] [信息] 总请求数: 3 (成功: 3, 失败: 0) +[2025-04-22 15:17:42.169] [信息] 超时比率: 0.00% +[2025-04-22 15:17:42.169] [信息] 平均响应时间: 22.476s +[2025-04-22 15:17:42.169] [信息] P90响应时间: 24.723s +[2025-04-22 15:17:42.169] [信息] P95响应时间: 24.723s +[2025-04-22 15:17:42.169] [信息] P99响应时间: 24.723s +[2025-04-22 15:17:42.169] [信息] 平均QPS: 0.06 +[2025-04-22 15:17:42.169] [信息] 最大QPS: 205128.21 +[2025-04-22 15:17:42.169] [信息] 平均Token生成速率: 19.25 tokens/s +[2025-04-22 15:17:42.169] [信息] 最大Token生成速率: 68307692.31 tokens/s +[2025-04-22 15:17:42.169] [信息] 最大有效并发用户数: 2 +[2025-04-22 15:17:42.169] [信息] 最小首个令牌生成时间: 9.267s +[2025-04-22 15:17:42.169] [信息] 最大首个令牌生成时间: 11.273s +[2025-04-22 15:17:42.169] [信息] P90首个令牌生成时间: 11.088s +[2025-04-22 15:17:42.169] [信息] P95首个令牌生成时间: 11.088s +[2025-04-22 15:17:42.169] [信息] P99首个令牌生成时间: 11.088s +[2025-04-22 15:17:42.169] [信息] 生成HTML报告... +[2025-04-22 15:17:42.169] [调试] 创建HTML报告页面... +[2025-04-22 15:17:42.169] [调试] 添加概述图表... +[2025-04-22 15:17:42.169] [调试] 添加响应时间分布图... +[2025-04-22 15:17:42.170] [调试] 添加QPS随时间变化图... +[2025-04-22 15:17:42.170] [调试] 添加Token生成速率随时间变化图... +[2025-04-22 15:17:42.170] [调试] 添加并发与响应时间关系图... +[2025-04-22 15:17:42.170] [调试] 添加TTFT随时间变化图... +[2025-04-22 15:17:42.170] [调试] 保存HTML报告: reports/report_2025-04-22_15-17-42/report.html +[2025-04-22 15:17:42.175] [信息] 生成CSV数据... +[2025-04-22 15:17:42.175] [调试] 创建CSV文件: reports/report_2025-04-22_15-17-42/results.csv +[2025-04-22 15:17:42.176] [调试] 写入CSV头... +[2025-04-22 15:17:42.176] [调试] 写入CSV数据,共3条记录... +[2025-04-22 15:17:42.176] [调试] CSV数据写入完成 +[2025-04-22 15:17:42.176] [信息] 报告生成完成 +[2025-04-22 15:17:42.176] [信息] 基准测试完成!报告已生成: reports/report_2025-04-22_15-17-42 +[2025-04-22 15:17:42.176] [信息] HTML报告: reports/report_2025-04-22_15-17-42/report.html +[2025-04-22 15:17:42.176] [信息] CSV数据: reports/report_2025-04-22_15-17-42/results.csv diff --git a/reports/report_2025-04-22_15-17-42/report.html b/reports/report_2025-04-22_15-17-42/report.html new file mode 100644 index 0000000..9eefc77 --- /dev/null +++ b/reports/report_2025-04-22_15-17-42/report.html @@ -0,0 +1,121 @@ + + + + + + LLM API 基准测试报告 + + + + + + + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + + + + + + diff --git a/reports/report_2025-04-22_15-17-42/results.csv b/reports/report_2025-04-22_15-17-42/results.csv new file mode 100644 index 0000000..b2f66d9 --- /dev/null +++ b/reports/report_2025-04-22_15-17-42/results.csv @@ -0,0 +1,4 @@ +Timestamp,PromptType,PromptTokens,CompletionTokens,ResponseTime,Concurrency,FirstTokenTime,Error +2025-04-22T15:16:51+08:00,short,50,314,24791,1,11273, +2025-04-22T15:17:17+08:00,short,49,216,17983,2,9267, +2025-04-22T15:17:17+08:00,short,47,450,24655,2,10903, diff --git a/reports/report_2025-04-22_16-39-44/api_logs.txt b/reports/report_2025-04-22_16-39-44/api_logs.txt new file mode 100644 index 0000000..d91cab1 --- /dev/null +++ b/reports/report_2025-04-22_16-39-44/api_logs.txt @@ -0,0 +1,55 @@ +[2025-04-22 16:39:44.748] [信息] 日志文件已设置: reports/report_2025-04-22_16-39-44/api_logs.txt +[2025-04-22 16:39:44.748] [信息] API请求响应日志将记录到: reports/report_2025-04-22_16-39-44/api_logs.txt +[2025-04-22 16:39:44.749] [信息] 创建报告输出目录: reports/report_2025-04-22_16-39-44 +[2025-04-22 16:39:44.749] [调试] 共有 3 条测试结果记录 +[2025-04-22 16:39:44.749] [信息] 计算性能指标... +[2025-04-22 16:39:44.749] [调试] 计算指标 - 总请求数: 3 +[2025-04-22 16:39:44.749] [调试] 计算指标 - 成功请求: 3, 失败请求: 0 +[2025-04-22 16:39:44.749] [调试] 计算指标 - 超时比率: 0.00% +[2025-04-22 16:39:44.749] [调试] 计算指标 - 响应时间(ms) - 平均: 34273.33, P90: 38649.50, P95: 38649.50, P99: 38649.50, 最小: 25521.00, 最大: 48694.00 +[2025-04-22 16:39:44.749] [调试] 计算指标 - TTFT(ms) - 最小: 11289.00, 最大: 14450.00, P90: 13680.00, P95: 13680.00, P99: 13680.00 +[2025-04-22 16:39:44.749] [调试] 计算指标 - 平均QPS: 0.04 (总持续时间: 77.80秒) +[2025-04-22 16:39:44.749] [调试] 计算各并发步骤的QPS... +[2025-04-22 16:39:44.749] [调试] 并发步骤 1: 持续时间过短 +[2025-04-22 16:39:44.749] [调试] 并发步骤 2: QPS = 431.68 (成功请求: 2, 持续时间: 0.00秒) +[2025-04-22 16:39:44.749] [调试] 更新最大QPS: 431.68 (并发步骤: 2) +[2025-04-22 16:39:44.749] [调试] 更新最大有效并发用户数: 2 +[2025-04-22 16:39:44.749] [调试] 计算指标 - 平均Token生成速率: 21.53 tokens/s (总Token: 1675) +[2025-04-22 16:39:44.749] [调试] 计算各并发步骤的Token生成速率... +[2025-04-22 16:39:44.749] [调试] 并发步骤 2: Token生成速率 = 275628.84 tokens/s (总Token: 1277, 持续时间: 0.00秒) +[2025-04-22 16:39:44.749] [调试] 更新最大Token生成速率: 275628.84 tokens/s (并发步骤: 2) +[2025-04-22 16:39:44.749] [信息] 性能指标摘要: +[2025-04-22 16:39:44.749] [信息] 总请求数: 3 (成功: 3, 失败: 0) +[2025-04-22 16:39:44.749] [信息] 超时比率: 0.00% +[2025-04-22 16:39:44.749] [信息] 平均响应时间: 34.273s +[2025-04-22 16:39:44.749] [信息] P90响应时间: 38.649s +[2025-04-22 16:39:44.749] [信息] P95响应时间: 38.649s +[2025-04-22 16:39:44.749] [信息] P99响应时间: 38.649s +[2025-04-22 16:39:44.749] [信息] 平均QPS: 0.04 +[2025-04-22 16:39:44.749] [信息] 最大QPS: 431.68 +[2025-04-22 16:39:44.749] [信息] 平均Token生成速率: 21.53 tokens/s +[2025-04-22 16:39:44.749] [信息] 最大Token生成速率: 275628.84 tokens/s +[2025-04-22 16:39:44.749] [信息] 最大有效并发用户数: 2 +[2025-04-22 16:39:44.749] [信息] 最小首个令牌生成时间: 11.289s +[2025-04-22 16:39:44.749] [信息] 最大首个令牌生成时间: 14.45s +[2025-04-22 16:39:44.749] [信息] P90首个令牌生成时间: 13.68s +[2025-04-22 16:39:44.749] [信息] P95首个令牌生成时间: 13.68s +[2025-04-22 16:39:44.749] [信息] P99首个令牌生成时间: 13.68s +[2025-04-22 16:39:44.749] [信息] 生成HTML报告... +[2025-04-22 16:39:44.750] [调试] 创建HTML报告页面... +[2025-04-22 16:39:44.750] [调试] 添加概述图表... +[2025-04-22 16:39:44.750] [调试] 添加响应时间分布图... +[2025-04-22 16:39:44.750] [调试] 添加QPS随时间变化图... +[2025-04-22 16:39:44.750] [调试] 添加Token生成速率随时间变化图... +[2025-04-22 16:39:44.750] [调试] 添加并发与响应时间关系图... +[2025-04-22 16:39:44.750] [调试] 添加TTFT随时间变化图... +[2025-04-22 16:39:44.751] [调试] 保存HTML报告: reports/report_2025-04-22_16-39-44/report.html +[2025-04-22 16:39:44.756] [信息] 生成CSV数据... +[2025-04-22 16:39:44.757] [调试] 创建CSV文件: reports/report_2025-04-22_16-39-44/results.csv +[2025-04-22 16:39:44.757] [调试] 写入CSV头... +[2025-04-22 16:39:44.757] [调试] 写入CSV数据,共3条记录... +[2025-04-22 16:39:44.757] [调试] CSV数据写入完成 +[2025-04-22 16:39:44.757] [信息] 报告生成完成 +[2025-04-22 16:39:44.757] [信息] 基准测试完成!报告已生成: reports/report_2025-04-22_16-39-44 +[2025-04-22 16:39:44.757] [信息] HTML报告: reports/report_2025-04-22_16-39-44/report.html +[2025-04-22 16:39:44.757] [信息] CSV数据: reports/report_2025-04-22_16-39-44/results.csv diff --git a/reports/report_2025-04-22_16-39-44/report.html b/reports/report_2025-04-22_16-39-44/report.html new file mode 100644 index 0000000..e658971 --- /dev/null +++ b/reports/report_2025-04-22_16-39-44/report.html @@ -0,0 +1,121 @@ + + + + + + LLM API 基准测试报告 + + + + + + + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + + + + + + diff --git a/reports/report_2025-04-22_16-39-44/results.csv b/reports/report_2025-04-22_16-39-44/results.csv new file mode 100644 index 0000000..873c534 --- /dev/null +++ b/reports/report_2025-04-22_16-39-44/results.csv @@ -0,0 +1,4 @@ +Timestamp,PromptType,PromptTokens,CompletionTokens,ResponseTime,Concurrency,FirstTokenTime,Error +2025-04-22T16:38:26+08:00,short,49,398,28605,1,12910, +2025-04-22T16:38:55+08:00,short,49,371,25521,2,11289, +2025-04-22T16:38:55+08:00,long,1005,906,48694,2,14450, diff --git a/reports/report_2025-04-22_17-16-42/api_logs.txt b/reports/report_2025-04-22_17-16-42/api_logs.txt new file mode 100644 index 0000000..ced3a8d --- /dev/null +++ b/reports/report_2025-04-22_17-16-42/api_logs.txt @@ -0,0 +1,44 @@ +[2025-04-22 17:16:42.378] [信息] 日志文件已设置: reports/report_2025-04-22_17-16-42/api_logs.txt +[2025-04-22 17:16:42.378] [信息] API请求响应日志将记录到: reports/report_2025-04-22_17-16-42/api_logs.txt +[2025-04-22 17:16:42.379] [信息] 创建报告输出目录: reports/report_2025-04-22_17-16-42 +[2025-04-22 17:16:42.380] [调试] 共有 3 条测试结果记录 +[2025-04-22 17:16:42.380] [信息] 计算性能指标... +[2025-04-22 17:16:42.380] [调试] 计算指标 - 总请求数: 3 +[2025-04-22 17:16:42.380] [调试] 计算指标 - 成功请求: 0, 失败请求: 3 +[2025-04-22 17:16:42.380] [调试] 计算指标 - 超时比率: 100.00% +[2025-04-22 17:16:42.380] [警告] 没有成功的请求,无法计算详细指标 +[2025-04-22 17:16:42.380] [信息] 性能指标摘要: +[2025-04-22 17:16:42.380] [信息] 总请求数: 3 (成功: 0, 失败: 3) +[2025-04-22 17:16:42.380] [信息] 超时比率: 100.00% +[2025-04-22 17:16:42.380] [信息] 平均响应时间: 0s +[2025-04-22 17:16:42.380] [信息] P90响应时间: 0s +[2025-04-22 17:16:42.380] [信息] P95响应时间: 0s +[2025-04-22 17:16:42.380] [信息] P99响应时间: 0s +[2025-04-22 17:16:42.380] [信息] 平均QPS: 0.00 +[2025-04-22 17:16:42.380] [信息] 最大QPS: 0.00 +[2025-04-22 17:16:42.380] [信息] 平均Token生成速率: 0.00 tokens/s +[2025-04-22 17:16:42.380] [信息] 最大Token生成速率: 0.00 tokens/s +[2025-04-22 17:16:42.380] [信息] 最大有效并发用户数: 0 +[2025-04-22 17:16:42.380] [信息] 最小首个令牌生成时间: 0s +[2025-04-22 17:16:42.380] [信息] 最大首个令牌生成时间: 0s +[2025-04-22 17:16:42.380] [信息] P90首个令牌生成时间: 0s +[2025-04-22 17:16:42.380] [信息] P95首个令牌生成时间: 0s +[2025-04-22 17:16:42.380] [信息] P99首个令牌生成时间: 0s +[2025-04-22 17:16:42.380] [信息] 生成HTML报告... +[2025-04-22 17:16:42.380] [调试] 创建HTML报告页面... +[2025-04-22 17:16:42.380] [调试] 添加概述图表... +[2025-04-22 17:16:42.380] [调试] 添加响应时间分布图... +[2025-04-22 17:16:42.380] [调试] 添加QPS随时间变化图... +[2025-04-22 17:16:42.381] [调试] 添加Token生成速率随时间变化图... +[2025-04-22 17:16:42.381] [调试] 添加并发与响应时间关系图... +[2025-04-22 17:16:42.381] [调试] 添加TTFT随时间变化图... +[2025-04-22 17:16:42.381] [调试] 保存HTML报告: reports/report_2025-04-22_17-16-42/report.html +[2025-04-22 17:16:42.389] [信息] 生成CSV数据... +[2025-04-22 17:16:42.389] [调试] 创建CSV文件: reports/report_2025-04-22_17-16-42/results.csv +[2025-04-22 17:16:42.389] [调试] 写入CSV头... +[2025-04-22 17:16:42.390] [调试] 写入CSV数据,共3条记录... +[2025-04-22 17:16:42.390] [调试] CSV数据写入完成 +[2025-04-22 17:16:42.390] [信息] 报告生成完成 +[2025-04-22 17:16:42.390] [信息] 基准测试完成!报告已生成: reports/report_2025-04-22_17-16-42 +[2025-04-22 17:16:42.390] [信息] HTML报告: reports/report_2025-04-22_17-16-42/report.html +[2025-04-22 17:16:42.390] [信息] CSV数据: reports/report_2025-04-22_17-16-42/results.csv diff --git a/reports/report_2025-04-22_17-16-42/report.html b/reports/report_2025-04-22_17-16-42/report.html new file mode 100644 index 0000000..1ca61a7 --- /dev/null +++ b/reports/report_2025-04-22_17-16-42/report.html @@ -0,0 +1,121 @@ + + + + + + LLM API 基准测试报告 + + + + + + + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + + + + + + diff --git a/reports/report_2025-04-22_17-16-42/results.csv b/reports/report_2025-04-22_17-16-42/results.csv new file mode 100644 index 0000000..7c09aac --- /dev/null +++ b/reports/report_2025-04-22_17-16-42/results.csv @@ -0,0 +1,4 @@ +Timestamp,PromptType,PromptTokens,CompletionTokens,ResponseTime,Concurrency,FirstTokenTime,Error +2025-04-22T17:15:36+08:00,long,979,0,0,1,0,发送请求失败: Post "http://139.9.229.200:8080/v1/chat/completions": dial tcp 139.9.229.200:8080: i/o timeout +2025-04-22T17:16:07+08:00,short,51,0,0,2,0,发送请求失败: Post "http://139.9.229.200:8080/v1/chat/completions": dial tcp 139.9.229.200:8080: i/o timeout +2025-04-22T17:16:07+08:00,long,955,0,0,2,0,发送请求失败: Post "http://139.9.229.200:8080/v1/chat/completions": dial tcp 139.9.229.200:8080: i/o timeout diff --git a/reports/report_2025-04-22_17-18-50/api_logs.txt b/reports/report_2025-04-22_17-18-50/api_logs.txt new file mode 100644 index 0000000..e05b4f7 --- /dev/null +++ b/reports/report_2025-04-22_17-18-50/api_logs.txt @@ -0,0 +1,55 @@ +[2025-04-22 17:18:50.907] [信息] 日志文件已设置: reports/report_2025-04-22_17-18-50/api_logs.txt +[2025-04-22 17:18:50.910] [信息] API请求响应日志将记录到: reports/report_2025-04-22_17-18-50/api_logs.txt +[2025-04-22 17:18:50.910] [信息] 创建报告输出目录: reports/report_2025-04-22_17-18-50 +[2025-04-22 17:18:50.910] [调试] 共有 3 条测试结果记录 +[2025-04-22 17:18:50.910] [信息] 计算性能指标... +[2025-04-22 17:18:50.910] [调试] 计算指标 - 总请求数: 3 +[2025-04-22 17:18:50.910] [调试] 计算指标 - 成功请求: 3, 失败请求: 0 +[2025-04-22 17:18:50.910] [调试] 计算指标 - 超时比率: 0.00% +[2025-04-22 17:18:50.910] [调试] 计算指标 - 响应时间(ms) - 平均: 41352.67, P90: 49964.00, P95: 49964.00, P99: 49964.00, 最小: 24130.00, 最大: 51447.00 +[2025-04-22 17:18:50.910] [调试] 计算指标 - TTFT(ms) - 最小: 11284.00, 最大: 14291.00, P90: 13733.50, P95: 13733.50, P99: 13733.50 +[2025-04-22 17:18:50.910] [调试] 计算指标 - 平均QPS: 0.03 (总持续时间: 100.52秒) +[2025-04-22 17:18:50.910] [调试] 计算各并发步骤的QPS... +[2025-04-22 17:18:50.910] [调试] 并发步骤 1: 持续时间过短 +[2025-04-22 17:18:50.910] [调试] 并发步骤 2: QPS = 458.27 (成功请求: 2, 持续时间: 0.00秒) +[2025-04-22 17:18:50.910] [调试] 更新最大QPS: 458.27 (并发步骤: 2) +[2025-04-22 17:18:50.910] [调试] 更新最大有效并发用户数: 2 +[2025-04-22 17:18:50.910] [调试] 计算指标 - 平均Token生成速率: 23.57 tokens/s (总Token: 2369) +[2025-04-22 17:18:50.910] [调试] 计算各并发步骤的Token生成速率... +[2025-04-22 17:18:50.910] [调试] 并发步骤 2: Token生成速率 = 286418.06 tokens/s (总Token: 1250, 持续时间: 0.00秒) +[2025-04-22 17:18:50.910] [调试] 更新最大Token生成速率: 286418.06 tokens/s (并发步骤: 2) +[2025-04-22 17:18:50.910] [信息] 性能指标摘要: +[2025-04-22 17:18:50.910] [信息] 总请求数: 3 (成功: 3, 失败: 0) +[2025-04-22 17:18:50.910] [信息] 超时比率: 0.00% +[2025-04-22 17:18:50.910] [信息] 平均响应时间: 41.352s +[2025-04-22 17:18:50.910] [信息] P90响应时间: 49.964s +[2025-04-22 17:18:50.910] [信息] P95响应时间: 49.964s +[2025-04-22 17:18:50.910] [信息] P99响应时间: 49.964s +[2025-04-22 17:18:50.910] [信息] 平均QPS: 0.03 +[2025-04-22 17:18:50.910] [信息] 最大QPS: 458.27 +[2025-04-22 17:18:50.910] [信息] 平均Token生成速率: 23.57 tokens/s +[2025-04-22 17:18:50.910] [信息] 最大Token生成速率: 286418.06 tokens/s +[2025-04-22 17:18:50.910] [信息] 最大有效并发用户数: 2 +[2025-04-22 17:18:50.910] [信息] 最小首个令牌生成时间: 11.284s +[2025-04-22 17:18:50.910] [信息] 最大首个令牌生成时间: 14.291s +[2025-04-22 17:18:50.910] [信息] P90首个令牌生成时间: 13.733s +[2025-04-22 17:18:50.910] [信息] P95首个令牌生成时间: 13.733s +[2025-04-22 17:18:50.910] [信息] P99首个令牌生成时间: 13.733s +[2025-04-22 17:18:50.910] [信息] 生成HTML报告... +[2025-04-22 17:18:50.910] [调试] 创建HTML报告页面... +[2025-04-22 17:18:50.910] [调试] 添加概述图表... +[2025-04-22 17:18:50.910] [调试] 添加响应时间分布图... +[2025-04-22 17:18:50.910] [调试] 添加QPS随时间变化图... +[2025-04-22 17:18:50.910] [调试] 添加Token生成速率随时间变化图... +[2025-04-22 17:18:50.910] [调试] 添加并发与响应时间关系图... +[2025-04-22 17:18:50.911] [调试] 添加TTFT随时间变化图... +[2025-04-22 17:18:50.911] [调试] 保存HTML报告: reports/report_2025-04-22_17-18-50/report.html +[2025-04-22 17:18:50.914] [信息] 生成CSV数据... +[2025-04-22 17:18:50.914] [调试] 创建CSV文件: reports/report_2025-04-22_17-18-50/results.csv +[2025-04-22 17:18:50.915] [调试] 写入CSV头... +[2025-04-22 17:18:50.915] [调试] 写入CSV数据,共3条记录... +[2025-04-22 17:18:50.915] [调试] CSV数据写入完成 +[2025-04-22 17:18:50.915] [信息] 报告生成完成 +[2025-04-22 17:18:50.915] [信息] 基准测试完成!报告已生成: reports/report_2025-04-22_17-18-50 +[2025-04-22 17:18:50.915] [信息] HTML报告: reports/report_2025-04-22_17-18-50/report.html +[2025-04-22 17:18:50.915] [信息] CSV数据: reports/report_2025-04-22_17-18-50/results.csv diff --git a/reports/report_2025-04-22_17-18-50/report.html b/reports/report_2025-04-22_17-18-50/report.html new file mode 100644 index 0000000..daf5772 --- /dev/null +++ b/reports/report_2025-04-22_17-18-50/report.html @@ -0,0 +1,121 @@ + + + + + + LLM API 基准测试报告 + + + + + + + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + + + + + + diff --git a/reports/report_2025-04-22_17-18-50/results.csv b/reports/report_2025-04-22_17-18-50/results.csv new file mode 100644 index 0000000..bd1c494 --- /dev/null +++ b/reports/report_2025-04-22_17-18-50/results.csv @@ -0,0 +1,4 @@ +Timestamp,PromptType,PromptTokens,CompletionTokens,ResponseTime,Concurrency,FirstTokenTime,Error +2025-04-22T17:17:10+08:00,long,958,1119,51447,1,13176, +2025-04-22T17:18:01+08:00,short,50,250,24130,2,14291, +2025-04-22T17:18:01+08:00,long,957,1000,48481,2,11284, diff --git a/reports/report_2025-04-22_17-37-34/api_logs.txt b/reports/report_2025-04-22_17-37-34/api_logs.txt new file mode 100644 index 0000000..7a0d2b4 --- /dev/null +++ b/reports/report_2025-04-22_17-37-34/api_logs.txt @@ -0,0 +1,55 @@ +[2025-04-22 17:37:34.946] [信息] 日志文件已设置: reports/report_2025-04-22_17-37-34/api_logs.txt +[2025-04-22 17:37:34.946] [信息] API请求响应日志将记录到: reports/report_2025-04-22_17-37-34/api_logs.txt +[2025-04-22 17:37:34.946] [信息] 创建报告输出目录: reports/report_2025-04-22_17-37-34 +[2025-04-22 17:37:34.946] [调试] 共有 3 条测试结果记录 +[2025-04-22 17:37:34.946] [信息] 计算性能指标... +[2025-04-22 17:37:34.946] [调试] 计算指标 - 总请求数: 3 +[2025-04-22 17:37:34.946] [调试] 计算指标 - 成功请求: 3, 失败请求: 0 +[2025-04-22 17:37:34.946] [调试] 计算指标 - 超时比率: 0.00% +[2025-04-22 17:37:34.946] [调试] 计算指标 - 响应时间(ms) - 平均: 48196.33, P90: 54100.50, P95: 54100.50, P99: 54100.50, 最小: 36388.00, 最大: 54352.00 +[2025-04-22 17:37:34.946] [调试] 计算指标 - TTFT(ms) - 最小: 12860.00, 最大: 23094.00, P90: 20190.00, P95: 20190.00, P99: 20190.00 +[2025-04-22 17:37:34.946] [调试] 计算指标 - 平均QPS: 0.03 (总持续时间: 110.94秒) +[2025-04-22 17:37:34.946] [调试] 计算各并发步骤的QPS... +[2025-04-22 17:37:34.946] [调试] 并发步骤 1: 持续时间过短 +[2025-04-22 17:37:34.946] [调试] 并发步骤 2: QPS = 968.00 (成功请求: 2, 持续时间: 0.00秒) +[2025-04-22 17:37:34.946] [调试] 更新最大QPS: 968.00 (并发步骤: 2) +[2025-04-22 17:37:34.946] [调试] 更新最大有效并发用户数: 2 +[2025-04-22 17:37:34.946] [调试] 计算指标 - 平均Token生成速率: 27.46 tokens/s (总Token: 3046) +[2025-04-22 17:37:34.946] [调试] 计算各并发步骤的Token生成速率... +[2025-04-22 17:37:34.946] [调试] 并发步骤 2: Token生成速率 = 884748.02 tokens/s (总Token: 1828, 持续时间: 0.00秒) +[2025-04-22 17:37:34.946] [调试] 更新最大Token生成速率: 884748.02 tokens/s (并发步骤: 2) +[2025-04-22 17:37:34.946] [信息] 性能指标摘要: +[2025-04-22 17:37:34.946] [信息] 总请求数: 3 (成功: 3, 失败: 0) +[2025-04-22 17:37:34.946] [信息] 超时比率: 0.00% +[2025-04-22 17:37:34.946] [信息] 平均响应时间: 48.196s +[2025-04-22 17:37:34.946] [信息] P90响应时间: 54.1s +[2025-04-22 17:37:34.946] [信息] P95响应时间: 54.1s +[2025-04-22 17:37:34.946] [信息] P99响应时间: 54.1s +[2025-04-22 17:37:34.946] [信息] 平均QPS: 0.03 +[2025-04-22 17:37:34.946] [信息] 最大QPS: 968.00 +[2025-04-22 17:37:34.946] [信息] 平均Token生成速率: 27.46 tokens/s +[2025-04-22 17:37:34.946] [信息] 最大Token生成速率: 884748.02 tokens/s +[2025-04-22 17:37:34.946] [信息] 最大有效并发用户数: 2 +[2025-04-22 17:37:34.946] [信息] 最小首个令牌生成时间: 12.86s +[2025-04-22 17:37:34.946] [信息] 最大首个令牌生成时间: 23.094s +[2025-04-22 17:37:34.946] [信息] P90首个令牌生成时间: 20.19s +[2025-04-22 17:37:34.947] [信息] P95首个令牌生成时间: 20.19s +[2025-04-22 17:37:34.947] [信息] P99首个令牌生成时间: 20.19s +[2025-04-22 17:37:34.947] [信息] 生成HTML报告... +[2025-04-22 17:37:34.947] [调试] 创建HTML报告页面... +[2025-04-22 17:37:34.947] [调试] 添加概述图表... +[2025-04-22 17:37:34.947] [调试] 添加响应时间分布图... +[2025-04-22 17:37:34.947] [调试] 添加QPS随时间变化图... +[2025-04-22 17:37:34.947] [调试] 添加Token生成速率随时间变化图... +[2025-04-22 17:37:34.947] [调试] 添加并发与响应时间关系图... +[2025-04-22 17:37:34.947] [调试] 添加TTFT随时间变化图... +[2025-04-22 17:37:34.947] [调试] 保存HTML报告: reports/report_2025-04-22_17-37-34/report.html +[2025-04-22 17:37:34.953] [信息] 生成CSV数据... +[2025-04-22 17:37:34.953] [调试] 创建CSV文件: reports/report_2025-04-22_17-37-34/results.csv +[2025-04-22 17:37:34.953] [调试] 写入CSV头... +[2025-04-22 17:37:34.953] [调试] 写入CSV数据,共3条记录... +[2025-04-22 17:37:34.953] [调试] CSV数据写入完成 +[2025-04-22 17:37:34.953] [信息] 报告生成完成 +[2025-04-22 17:37:34.953] [信息] 基准测试完成!报告已生成: reports/report_2025-04-22_17-37-34 +[2025-04-22 17:37:34.953] [信息] HTML报告: reports/report_2025-04-22_17-37-34/report.html +[2025-04-22 17:37:34.953] [信息] CSV数据: reports/report_2025-04-22_17-37-34/results.csv diff --git a/reports/report_2025-04-22_17-37-34/report.html b/reports/report_2025-04-22_17-37-34/report.html new file mode 100644 index 0000000..5eb673d --- /dev/null +++ b/reports/report_2025-04-22_17-37-34/report.html @@ -0,0 +1,121 @@ + + + + + + LLM API 基准测试报告 + + + + + + + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + +
+
+
+ + + + + + + + diff --git a/reports/report_2025-04-22_17-37-34/results.csv b/reports/report_2025-04-22_17-37-34/results.csv new file mode 100644 index 0000000..d9fc670 --- /dev/null +++ b/reports/report_2025-04-22_17-37-34/results.csv @@ -0,0 +1,4 @@ +Timestamp,PromptType,PromptTokens,CompletionTokens,TotalTokens,ResponseTime(ms),FirstTokenTime(ms),Concurrency,Error,ReasoningContent +2025-04-22T17:35:44+08:00,long,973,1218,2191,53849,17286,1,,"" +2025-04-22T17:36:40+08:00,short,47,757,804,36388,12860,2,,"" +2025-04-22T17:36:40+08:00,long,970,1071,2041,54352,23094,2,,""