使用 Ollama 和 Qwen3 嵌入模型对文本文档进行重排序 - 使用 Go 语言

go build -o rnk

package main

import (
	"fmt"
	"log"
	"os"
	"sort"
	"time"

	"github.com/spf13/cobra"
)

var rootCmd = &cobra.Command{
	Use:   "rnk [query-file] [target-directory]",
	Short: "使用Ollama嵌入的RAG系统",
	Long:  "一个简单的RAG系统，提取嵌入并使用Ollama对文档进行排序",
	Args:  cobra.ExactArgs(2),
	Run:   runRnk,
}

var (
	embeddingModel string
	ollamaBaseURL  string
)

func init() {
	rootCmd.Flags().StringVarP(&embeddingModel, "model", "m", "dengcao/Qwen3-Embedding-4B:Q5_K_M", "使用的嵌入模型")
	rootCmd.Flags().StringVarP(&ollamaBaseURL, "url", "u", "http://localhost:11434", "Ollama基础URL")
}

func main() {
	if err := rootCmd.Execute(); err != nil {
		fmt.Println(err)
		os.Exit(1)
	}
}

func runRnk(cmd *cobra.Command, args []string) {
	queryFile := args[0]
	targetDir := args[1]

	startTime := time.Now()

	fmt.Printf("使用嵌入模型: %s\n", embeddingModel)
	fmt.Printf("Ollama基础URL: %s\n", ollamaBaseURL)
	fmt.Printf("处理查询文件: %s, 目标目录: %s\n", queryFile, targetDir)

	// 从文件读取查询
	query, err := readQueryFromFile(queryFile)
	if err != nil {
		log.Fatalf("读取查询文件时出错: %v", err)
	}
	fmt.Printf("查询: %s\n", query)

	// 查找目标目录中的所有文本文件
	documents, err := findTextFiles(targetDir)
	if err != nil {
		log.Fatalf("查找文本文件时出错: %v", err)
	}
	fmt.Printf("找到 %d 个文档\n", len(documents))

	// 提取查询的嵌入
	fmt.Println("提取查询嵌入...")
	queryEmbedding, err := getEmbedding(query, embeddingModel, ollamaBaseURL)
	if err != nil {
		log.Fatalf("获取查询嵌入时出错: %v", err)
	}

	// 处理文档
	fmt.Println("处理文档...")
	validDocs := make([]Document, 0)

	for _, doc := range documents {
		embedding, err := getEmbedding(doc.Content, embeddingModel, ollamaBaseURL)
		if err != nil {
			fmt.Printf("警告: 无法获取 %s 的嵌入: %v\n", doc.Path, err)
			continue
		}

		similarity := cosineSimilarity(queryEmbedding, embedding)
		doc.Score = similarity
		validDocs = append(validDocs, doc)
	}

	if len(validDocs) == 0 {
		log.Fatalf("没有文档可以成功处理")
	}

	// 按相似度分数降序排序
	sort.Slice(validDocs, func(i, j int) bool {
		return validDocs[i].Score > validDocs[j].Score
	})

	// 显示结果
	fmt.Println("\n=== 按相似度排序 ===")
	for i, doc := range validDocs {
		fmt.Printf("%d. %s (得分: %.3f)\n", i+1, doc.Path, doc.Score)
	}

	totalTime := time.Since(startTime)
	avgTimePerDoc := totalTime / time.Duration(len(validDocs))

	fmt.Printf("\n处理了 %d 个文档，耗时 %.3fs (平均: %.3fs 每个文档)\n",
		len(validDocs), totalTime.Seconds(), avgTimePerDoc.Seconds())
}

package main

import (
	"fmt"
	"os"
	"path/filepath"
	"strings"
)

func readQueryFromFile(filename string) (string, error) {
	content, err := os.ReadFile(filename)
	if err != nil {
		return "", err
	}
	return strings.TrimSpace(string(content)), nil
}

func findTextFiles(dir string) ([]Document, error) {
	var documents []Document

	err := filepath.Walk(dir, func(path string, info os.FileInfo, err error) error {
		if err != nil {
			return err
		}

		if !info.IsDir() && isTextFile(path) {
			content, err := os.ReadFile(path)
			if err != nil {
				fmt.Printf("警告: 无法读取文件 %s: %v\n", path, err)
				return nil
			}

			documents = append(documents, Document{
				Path:    path,
				Content: string(content),
			})
		}

		return nil
	})

	return documents, err
}

func isTextFile(filename string) bool {
	ext := strings.ToLower(filepath.Ext(filename))
	textExts := []string{".txt", ".md", ".rst", ".csv", ".json", ".xml", ".html", ".htm", ".log"}
	for _, textExt := range textExts {
		if ext == textExt {
			return true
		}
	}
	return false
}

package main

import (
	"bytes"
	"encoding/json"
	"fmt"
	"io"
	"net/http"
)

func getEmbedding(text string, model string, ollamaBaseURL string) ([]float64, error) {
	req := OllamaEmbeddingRequest{
		Model:  model,
		Prompt: text,
	}

	jsonData, err := json.Marshal(req)
	if err != nil {
		return nil, err
	}

	resp, err := http.Post(ollamaBaseURL+"/api/embeddings", "application/json", bytes.NewBuffer(jsonData))
	if err != nil {
		return nil, err
	}
	defer resp.Body.Close()

	if resp.StatusCode != http.StatusOK {
		body, _ := io.ReadAll(resp.Body)
		return nil, fmt.Errorf("ollama API错误: %s", string(body))
	}

	var embeddingResp OllamaEmbeddingResponse
	if err := json.NewDecoder(resp.Body).Decode(&embeddingResp); err != nil {
		return nil, err
	}

	return embeddingResp.Embedding, nil
}

package main

func cosineSimilarity(a, b []float64) float64 {
	if len(a) != len(b) {
		return 0
	}

	var dotProduct, normA, normB float64

	for i := range a {
		dotProduct += a[i] * b[i]
		normA += a[i] * a[i]
		normB += b[i] * b[i]
	}

	if normA == 0 || normB == 0 {
		return 0
	}

	return dotProduct / (sqrt(normA) * sqrt(normB))
}

func sqrt(x float64) float64 {
	if x == 0 {
		return 0
	}
	z := x
	for i := 0; i < 10; i++ {
		z = (z + x/z) / 2
	}
	return z
}

package main

// OllamaEmbeddingRequest 代表Ollama嵌入API的请求负载
type OllamaEmbeddingRequest struct {
	Model  string `json:"model"`
	Prompt string `json:"prompt"`
}

// OllamaEmbeddingResponse 代表Ollama嵌入API的响应
type OllamaEmbeddingResponse struct {
	Embedding []float64 `json:"embedding"`
}

// Document 代表一个带有元数据的文档
type Document struct {
	Path    string
	Content string
	Score   float64
}