Add OpenAI integration (#10)

* backup current work * Add AI functionality * update docker-compose.yml * update docker-compose.yaml file --------- Co-authored-by: Erikas <erkexzcx@users.noreply.github.com>
2024-08-08 13:17:11 +03:00 · 2024-08-08 13:17:11 +03:00 · f29471cb5d
commit f29471cb5d
parent 00775d3514
10 changed files with 337 additions and 45 deletions
--- a/benchmarks.go
+++ b/benchmarks.go
@ -225,6 +225,10 @@ func postBenchmarkCreate(c *gin.Context) {
 		return
 	}

+	if openaiClient != nil {
+		go generateSummary(&benchmark, csvFiles)
+	}
+
 	// Redirect to the newly created benchmark using GET request
 	c.Redirect(http.StatusSeeOther, fmt.Sprintf("/benchmark/%d", benchmark.ID))
 }
@ -366,6 +370,11 @@ func getBenchmark(c *gin.Context) {
 		return
 	}

+	if benchmark.AiSummary == "" && openaiClient != nil {
+		go generateSummary(&benchmark, benchmarkDatas)
+		benchmark.AiSummary = "AI summary is being generated... Refresh the page later."
+	}
+
 	c.HTML(http.StatusOK, "benchmark.tmpl", gin.H{
 		"activePage": "benchmark",
 		"username":   session.Get("Username"),
--- a/config.go
+++ b/config.go
@ -10,39 +10,38 @@ import (
 type Config struct {
 	Bind          string
 	DataDir       string
+	SessionSecret string

 	DiscordClientID     string
 	DiscordClientSecret string
 	DiscordRedirectURL  string
-	SessionSecret       string
+
+	OpenAIURL    string
+	OpenAIApiKey string
+	OpenAIModel  string

 	Version bool
 }

 func NewConfig() (*Config, error) {
-	// Define the flags
-	bind := flag.String("bind", "0.0.0.0:8080", "Bind address and port")
-	dataDir := flag.String("data-dir", "/data", "Path where data would be stored")
-	discordClientID := flag.String("discord-client-id", "", "Discord OAuth2 client ID (see https://discord.com/developers/applications)")
-	discordClientSecret := flag.String("discord-client-secret", "", "Discord OAuth2 client secret (see https://discord.com/developers/applications)")
-	discordRedirectURL := flag.String("discord-redirect-url", "", "Discord OAuth2 redirect URL (<scheme>://<domain>/login/callback)")
-	sessionSecret := flag.String("session-secret", "", "Session secret")
-	flagVersion := flag.Bool("version", false, "prints version of the application")
+	config := &Config{}
+
+	flag.StringVar(&config.Bind, "bind", "0.0.0.0:8080", "Bind address and port")
+	flag.StringVar(&config.DataDir, "data-dir", "/data", "Path where data would be stored")
+	flag.StringVar(&config.SessionSecret, "session-secret", "", "Session secret")
+
+	flag.StringVar(&config.DiscordClientID, "discord-client-id", "", "Discord OAuth2 client ID (see https://discord.com/developers/applications)")
+	flag.StringVar(&config.DiscordClientSecret, "discord-client-secret", "", "Discord OAuth2 client secret (see https://discord.com/developers/applications)")
+	flag.StringVar(&config.DiscordRedirectURL, "discord-redirect-url", "", "Discord OAuth2 redirect URL (<scheme>://<domain>/login/callback)")
+
+	flag.StringVar(&config.OpenAIURL, "openai-url", "https://api.openai.com/v1", "OpenAI API URL")
+	flag.StringVar(&config.OpenAIModel, "openai-model", "gpt-4o", "OpenAI model ID")
+	flag.StringVar(&config.OpenAIApiKey, "openai-api-key", "", "OpenAI API Key (leave empty to disable OpenAI integration)")
+
+	flag.BoolVar(&config.Version, "version", false, "prints version of the application")

 	envflag.Parse(envflag.WithPrefix("FS_"))

-	// Assign the parsed flag values to the Config struct
-	config := &Config{
-		Bind:                *bind,
-		DataDir:             *dataDir,
-		DiscordClientID:     *discordClientID,
-		DiscordClientSecret: *discordClientSecret,
-		DiscordRedirectURL:  *discordRedirectURL,
-		SessionSecret:       *sessionSecret,
-
-		Version: *flagVersion,
-	}
-
 	if config.Version {
 		return config, nil
 	}
@ -62,6 +61,14 @@ func NewConfig() (*Config, error) {
 	if config.SessionSecret == "" {
 		return nil, errors.New("missing session-secret argument")
 	}
+	if config.OpenAIApiKey != "" {
+		if config.OpenAIModel == "" {
+			return nil, errors.New("missing openai-model argument")
+		}
+		if config.OpenAIURL == "" {
+			return nil, errors.New("missing openai-url argument")
+		}
+	}

 	return config, nil
 }
--- a/docker-compose.yaml
+++ b/docker-compose.yaml
@ -10,8 +10,11 @@ services:
    environment:
      - FS_BIND=0.0.0.0:8080
      - FS_DATA_DIR=/data
+      - FS_SESSION_SECRET=xxxxxxxxxxxxxxxxxx
      - FS_DISCORD_CLIENT_ID=xxxxxxxxxxxxxxxxxx
      - FS_DISCORD_CLIENT_SECRET=xxxxxxxxxxxxxxxxxx
      - FS_DISCORD_REDIRECT_URL=<scheme>://<domain>/login/callback
-      - FS_SESSION_SECRET=xxxxxxxxxxxxxxxxxx
+      - FS_OPENAI_API_KEY=xxxxxxxxxxxxxxxxxx
+      # - FS_OPENAI_MODEL=gpt-4o
+      # - FS_OPENAI_URL=https://api.openai.com/v1
    restart: unless-stopped
--- a/go.mod
+++ b/go.mod
@ -40,6 +40,7 @@ require (
 	github.com/pelletier/go-toml/v2 v2.2.2 // indirect
 	github.com/ravener/discord-oauth2 v0.0.0-20230514095040-ae65713199b3
 	github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec // indirect
+	github.com/sashabaranov/go-openai v1.28.0
 	github.com/twitchyliquid64/golang-asm v0.15.1 // indirect
 	github.com/ugorji/go/codec v1.2.12 // indirect
 	golang.org/x/arch v0.8.0 // indirect
--- a/go.sum
+++ b/go.sum
@ -171,6 +171,8 @@ github.com/rogpeppe/go-internal v1.3.0/go.mod h1:M8bDsm7K2OlrFYOpmOWEs/qY81heoFR
 github.com/rs/xid v1.2.1/go.mod h1:+uKXf+4Djp6Md1KODXJxgGQPKngRmWyn10oCKFzNHOQ=
 github.com/rs/zerolog v1.13.0/go.mod h1:YbFCdg8HfsridGWAh22vktObvhZbQsZXe4/zB0OKkWU=
 github.com/rs/zerolog v1.15.0/go.mod h1:xYTKnLHcpfU2225ny5qZjxnj9NvkumZYjJHlAThCjNc=
+github.com/sashabaranov/go-openai v1.28.0 h1:WS9F9BriSvtHvknPQy2Oi3b+8zkmJdEXcycrWqrSicQ=
+github.com/sashabaranov/go-openai v1.28.0/go.mod h1:lj5b/K+zjTSFxVLijLSTDZuP7adOgerWeFyZLUhAKRg=
 github.com/satori/go.uuid v1.2.0/go.mod h1:dA0hQrYB0VpLJoorglMZABFdXlWrHn1NEOzdhQKdks0=
 github.com/shopspring/decimal v0.0.0-20180709203117-cd690d0c9e24/go.mod h1:M+9NzErvs504Cn4c5DxATwIqPbtswREoFCre64PpcG4=
 github.com/shopspring/decimal v1.2.0/go.mod h1:DKyhrW/HYNuLGql+MJL6WCR6knT2jwCFRcu2hWCYk4o=
--- a/models.go
+++ b/models.go
@ -18,6 +18,7 @@ type Benchmark struct {
 	UserID      uint
 	Title       string
 	Description string
+	AiSummary   string

 	CreatedAtHumanized string `gorm:"-"` // Human readable "X h/m/s ago" version of CreatedAt (filled automatically)

--- a/openai.go
+++ b/openai.go
@ -0,0 +1,270 @@
+package flightlesssomething
+
+import (
+	"context"
+	"log"
+	"math"
+	"sort"
+	"strconv"
+	"strings"
+	"sync"
+
+	openai "github.com/sashabaranov/go-openai"
+)
+
+const systemMessage = `
+You are given PC benchmark data of several runs. All this data is visible in the website in a form of charts and your goal is to provide insights.
+
+You MUST:
+1. Write at max 3 sections (headers) - "Top runs", "Issues" (optional) and "Summary".
+2. In Issues section, Figure out if any of the run is significantly worse then others in the same benchmark. You MUST use ONLY the data provided to explain the difference, and your points must be based only on the data provided. If there are no issues - do not write this section. Do not make any guesses. Additional requirements: (a) validate if the same hardware/software was used (by using provided text fields, NOT the data), (b) do not speculate, but use numbers to back up your claims, (c) only write if it's an actual issue with FPS (everything else is just additional information).
+3. In Top runs section, provide which run has the (average) "Highest FPS", which has the "Smoothest FPS" (LOWEST std.dev. and variance of FPS value - LOWEST, NOT HIGHEST) and which is the best "Best overall" (preferrably lower std.dev./variance than higher FPS, but if slight decrease in stability gives significantly higher FPS - pick that one). NEVER consider runs that have significantly lower FPS or has other significant issues. Exclude runs from consideration if they are significantly worse than the rest (as it would be mentioned in issues section). Note that your goal is to pick winners and not do a comparison in this section. Include numbers to justify your claims.
+4. In Summary section, provide an overview of all runs. Mention which runs are similar and which are different. Mention which runs are better in terms of FPS and which are better in terms of stability. Mention if there are any issues and what could be the reason for them. In short - summarize whole benchmark.
+5. First 2 sections should be bullet points, no subpoints, only 1 bullet point per point, while summary should be a single paragraph.
+6. NEVER use actual numbers. Instead, use percentage in comparison to other runs.
+7. Use markdown, use code syntax for labels.
+`
+
+var (
+	inProgressSummaries    = map[uint]struct{}{}
+	inProgressSummariesMux = &sync.Mutex{}
+)
+
+func generateSummary(b *Benchmark, bds []*BenchmarkData) {
+	// Check if OpenAI integration is not enabled
+	if openaiClient == nil {
+		return
+	}
+
+	// Lock mutex, as integration is enabled and might be already in progress
+	inProgressSummariesMux.Lock()
+
+	// Check if generation is already in progress
+	if _, ok := inProgressSummaries[b.ID]; ok {
+		inProgressSummariesMux.Unlock()
+		return
+	}
+	inProgressSummaries[b.ID] = struct{}{}
+	inProgressSummariesMux.Unlock()
+
+	// Create user prompt
+	userPrompt := writeAIPrompt(bds, b.Title, b.Description)
+
+	// Retrieve AI response
+	resp, err := openaiClient.CreateChatCompletion(
+		context.Background(),
+		openai.ChatCompletionRequest{
+			Model:       openaiModel,
+			Temperature: 0.0,
+			Messages: []openai.ChatCompletionMessage{
+				{Role: openai.ChatMessageRoleSystem, Content: systemMessage},
+				{Role: openai.ChatMessageRoleUser, Content: userPrompt},
+			},
+		},
+	)
+	if err != nil {
+		log.Println("Failed to generate AI summary:", err)
+		return
+	}
+
+	db.Model(&Benchmark{}).Where("id = ?", b.ID).Update("AiSummary", resp.Choices[0].Message.Content)
+
+	// Update status
+	inProgressSummariesMux.Lock()
+	delete(inProgressSummaries, b.ID)
+	inProgressSummariesMux.Unlock()
+}
+
+func writeAIPrompt(bds []*BenchmarkData, bdTitle, bdDescription string) string {
+	sb := strings.Builder{}
+	sb.WriteString("Benchmark title: ")
+	sb.WriteString(bdTitle)
+	sb.WriteString("\n")
+	sb.WriteString("Benchmark description: \n")
+	sb.WriteString(bdDescription)
+	sb.WriteString("\n\n")
+	sb.WriteString("Benchmark contains ")
+	sb.WriteString(strconv.Itoa(len(bds)))
+	sb.WriteString(" runs:\n")
+
+	for _, benchmarkRun := range bds {
+		sb.WriteString("\nLabel: ")
+		sb.WriteString(benchmarkRun.Label)
+		sb.WriteString("\n")
+
+		sb.WriteString("OS: ")
+		sb.WriteString(benchmarkRun.SpecOS)
+		sb.WriteString("\n")
+		sb.WriteString("GPU: ")
+		sb.WriteString(benchmarkRun.SpecGPU)
+		sb.WriteString("\n")
+		sb.WriteString("CPU: ")
+		sb.WriteString(benchmarkRun.SpecCPU)
+		sb.WriteString("\n")
+		sb.WriteString("RAM: ")
+		sb.WriteString(benchmarkRun.SpecRAM)
+		sb.WriteString("\n")
+		sb.WriteString("Linux kernel: ")
+		sb.WriteString(benchmarkRun.SpecLinuxKernel)
+		sb.WriteString("\n")
+		sb.WriteString("Linux scheduler: ")
+		sb.WriteString(benchmarkRun.SpecLinuxScheduler)
+		sb.WriteString("\n")
+
+		// FPS
+		stats := calculateAIPromptArrayStats(benchmarkRun.DataFPS)
+		sb.WriteString("FPS: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// Frame time
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataFrameTime)
+		sb.WriteString("Frame time: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// CPU load
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataCPULoad)
+		sb.WriteString("CPU load: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// GPU load
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataGPULoad)
+		sb.WriteString("GPU load: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// CPU temp
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataCPUTemp)
+		sb.WriteString("CPU temp: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// GPU temp
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataGPUTemp)
+		sb.WriteString("GPU temp: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// GPU core clock
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataGPUCoreClock)
+		sb.WriteString("GPU core clock: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// GPU mem clock
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataGPUMemClock)
+		sb.WriteString("GPU mem clock: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// GPU VRAM used
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataGPUVRAMUsed)
+		sb.WriteString("GPU VRAM used: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// GPU power
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataGPUPower)
+		sb.WriteString("GPU power: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// RAM used
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataRAMUsed)
+		sb.WriteString("RAM used: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+
+		// Swap used
+		stats = calculateAIPromptArrayStats(benchmarkRun.DataSwapUsed)
+		sb.WriteString("Swap used: ")
+		sb.WriteString(stats.String())
+		sb.WriteString("\n")
+	}
+
+	return sb.String()
+}
+
+type AIPromptArrayStats struct {
+	Count        int
+	Lowest       float64
+	Low1Percent  float64
+	Mean         float64
+	Median       float64
+	Top97Percent float64
+	Highest      float64
+	StdDev       float64
+	Variance     float64
+}
+
+func calculateAIPromptArrayStats(data []float64) AIPromptArrayStats {
+	if len(data) == 0 {
+		return AIPromptArrayStats{}
+	}
+
+	sort.Float64s(data)
+	count := len(data)
+	lowest := data[0]
+	highest := data[count-1]
+
+	low1PercentIndex := int(math.Ceil(0.01*float64(count))) - 1
+	if low1PercentIndex < 0 {
+		low1PercentIndex = 0
+	}
+	low1Percent := data[low1PercentIndex]
+
+	top97PercentIndex := int(math.Ceil(0.97*float64(count))) - 1
+	if top97PercentIndex < 0 {
+		top97PercentIndex = 0
+	}
+	top97Percent := data[top97PercentIndex]
+
+	mean := 0.0
+	for _, value := range data {
+		mean += value
+	}
+	mean /= float64(count)
+
+	median := 0.0
+	if count%2 == 0 {
+		median = (data[count/2-1] + data[count/2]) / 2
+	} else {
+		median = data[count/2]
+	}
+
+	variance := 0.0
+	for _, value := range data {
+		variance += (value - mean) * (value - mean)
+	}
+	variance /= float64(count)
+
+	stdDev := math.Sqrt(variance)
+
+	return AIPromptArrayStats{
+		Count:        count,
+		Lowest:       lowest,
+		Low1Percent:  low1Percent,
+		Mean:         mean,
+		Median:       median,
+		Top97Percent: top97Percent,
+		Highest:      highest,
+		StdDev:       stdDev,
+		Variance:     variance,
+	}
+}
+
+func (as AIPromptArrayStats) String() string {
+	return strings.Join([]string{
+		"Count: " + strconv.Itoa(as.Count),
+		"Lowest: " + strconv.FormatFloat(as.Lowest, 'f', -1, 64),
+		"Low1Percent: " + strconv.FormatFloat(as.Low1Percent, 'f', -1, 64),
+		"Mean: " + strconv.FormatFloat(as.Mean, 'f', -1, 64),
+		"Median: " + strconv.FormatFloat(as.Median, 'f', -1, 64),
+		"Top97Percent: " + strconv.FormatFloat(as.Top97Percent, 'f', -1, 64),
+		"Highest: " + strconv.FormatFloat(as.Highest, 'f', -1, 64),
+		"StdDev: " + strconv.FormatFloat(as.StdDev, 'f', -1, 64),
+		"Variance: " + strconv.FormatFloat(as.Variance, 'f', -1, 64),
+	}, ", ")
+}
--- a/server.go
+++ b/server.go
@ -15,6 +15,7 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/glebarez/sqlite"
 	"github.com/ravener/discord-oauth2"
+	openai "github.com/sashabaranov/go-openai"
 	"golang.org/x/oauth2"
 	"gorm.io/gorm"
 )
@ -28,9 +29,22 @@ var (

 	// Benchmarks directory
 	benchmarksDir string
+
+	// OpenAI
+	openaiClient *openai.Client
+	openaiModel  string
 )

 func Start(c *Config, version string) {
+	// Setup OpenAI client //
+
+	if c.OpenAIApiKey != "" {
+		openaiClientConf := openai.DefaultConfig(c.OpenAIApiKey)
+		openaiClientConf.BaseURL = c.OpenAIURL
+		openaiClient = openai.NewClientWithConfig(openaiClientConf)
+		openaiModel = c.OpenAIModel
+	}
+
 	// Setup data dir //

 	_, err := os.Stat(c.DataDir)
--- a/static/js/benchmark.js
+++ b/static/js/benchmark.js
@ -1,22 +1,7 @@
-document.addEventListener('DOMContentLoaded', function () {
-    var converter = new showdown.Converter();
-    var aiSummaryHTMLElement = document.getElementById('aiSummaryMarkdown');
-
-    // Create an Intersection Observer
-    var observer = new IntersectionObserver(function (entries, observer) {
-        entries.forEach(function (entry) {
-            if (entry.isIntersecting) {
-                // Element is visible, render the Markdown
-                var htmlContents = converter.makeHtml(aiSummaryHTMLElement.innerText);
-                aiSummaryHTMLElement.innerHTML = htmlContents;
-                // Stop observing after rendering
-                observer.unobserve(aiSummaryHTMLElement);
-            }
-        });
-    }, { threshold: 1.0 });
-
-    // Start observing the element
-    observer.observe(aiSummaryHTMLElement);
+document.addEventListener('DOMContentLoaded', () => {
+    const converter = new showdown.Converter();
+    const element = document.getElementById('aiSummaryMarkdown');
+    element.innerHTML = converter.makeHtml(element.innerHTML);
 });

 // ===========================================================================================
--- a/templates/benchmark.tmpl
+++ b/templates/benchmark.tmpl
@ -87,8 +87,8 @@
  <!-- Summary tab -->
  <div class="tab-pane" id="summary" role="tabpanel" aria-labelledby="summary-tab">
        <div class="row">
-            <div class="col-md-12 text-center" id="aiSummaryMarkdown">
-                <!-- TODO: markdown placeholder for AI summary -->
+            <div class="col-md-12" id="aiSummaryMarkdown">
+{{ .benchmark.AiSummary }}
            </div>
        </div>