261 lines
6.9 KiB
Go
261 lines
6.9 KiB
Go
package rag
|
|
|
|
import (
|
|
"encoding/binary"
|
|
"fmt"
|
|
"gf-lt/models"
|
|
"gf-lt/storage"
|
|
"log/slog"
|
|
"sort"
|
|
"strings"
|
|
"unsafe"
|
|
|
|
"github.com/jmoiron/sqlx"
|
|
)
|
|
|
|
// VectorStorage handles storing and retrieving vectors from SQLite
|
|
type VectorStorage struct {
|
|
logger *slog.Logger
|
|
sqlxDB *sqlx.DB
|
|
store storage.FullRepo
|
|
}
|
|
|
|
func NewVectorStorage(logger *slog.Logger, store storage.FullRepo) *VectorStorage {
|
|
return &VectorStorage{
|
|
logger: logger,
|
|
sqlxDB: store.DB(), // Use the new DB() method
|
|
store: store,
|
|
}
|
|
}
|
|
|
|
// SerializeVector converts []float32 to binary blob
|
|
func SerializeVector(vec []float32) []byte {
|
|
buf := make([]byte, len(vec)*4) // 4 bytes per float32
|
|
for i, v := range vec {
|
|
binary.LittleEndian.PutUint32(buf[i*4:], mathFloat32bits(v))
|
|
}
|
|
return buf
|
|
}
|
|
|
|
// DeserializeVector converts binary blob back to []float32
|
|
func DeserializeVector(data []byte) []float32 {
|
|
count := len(data) / 4
|
|
vec := make([]float32, count)
|
|
for i := 0; i < count; i++ {
|
|
vec[i] = mathBitsToFloat32(binary.LittleEndian.Uint32(data[i*4:]))
|
|
}
|
|
return vec
|
|
}
|
|
|
|
// mathFloat32bits and mathBitsToFloat32 are helpers to convert between float32 and uint32
|
|
func mathFloat32bits(f float32) uint32 {
|
|
return binary.LittleEndian.Uint32((*(*[4]byte)(unsafe.Pointer(&f)))[:4])
|
|
}
|
|
|
|
func mathBitsToFloat32(b uint32) float32 {
|
|
return *(*float32)(unsafe.Pointer(&b))
|
|
}
|
|
|
|
// WriteVector stores an embedding vector in the database
|
|
func (vs *VectorStorage) WriteVector(row *models.VectorRow) error {
|
|
tableName, err := vs.getTableName(row.Embeddings)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// Serialize the embeddings to binary
|
|
serializedEmbeddings := SerializeVector(row.Embeddings)
|
|
query := fmt.Sprintf(
|
|
"INSERT INTO %s (embeddings, slug, raw_text, filename) VALUES (?, ?, ?, ?)",
|
|
tableName,
|
|
)
|
|
if _, err := vs.sqlxDB.Exec(query, serializedEmbeddings, row.Slug, row.RawText, row.FileName); err != nil {
|
|
vs.logger.Error("failed to write vector", "error", err, "slug", row.Slug)
|
|
return err
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// getTableName determines which table to use based on embedding size
|
|
func (vs *VectorStorage) getTableName(emb []float32) (string, error) {
|
|
size := len(emb)
|
|
|
|
// Check if we support this embedding size
|
|
supportedSizes := map[int]bool{
|
|
384: true,
|
|
768: true,
|
|
1024: true,
|
|
1536: true,
|
|
2048: true,
|
|
3072: true,
|
|
4096: true,
|
|
5120: true,
|
|
}
|
|
if supportedSizes[size] {
|
|
return fmt.Sprintf("embeddings_%d", size), nil
|
|
}
|
|
return "", fmt.Errorf("no table for embedding size of %d", size)
|
|
}
|
|
|
|
// SearchClosest finds vectors closest to the query vector using efficient cosine similarity calculation
|
|
func (vs *VectorStorage) SearchClosest(query []float32) ([]models.VectorRow, error) {
|
|
tableName, err := vs.getTableName(query)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// For better performance, instead of loading all vectors at once,
|
|
// we'll implement batching and potentially add L2 distance-based pre-filtering
|
|
// since cosine similarity is related to L2 distance for normalized vectors
|
|
|
|
querySQL := "SELECT embeddings, slug, raw_text, filename FROM " + tableName
|
|
rows, err := vs.sqlxDB.Query(querySQL)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
defer rows.Close()
|
|
|
|
// Use a min-heap or simple slice to keep track of top 3 closest vectors
|
|
type SearchResult struct {
|
|
vector models.VectorRow
|
|
distance float32
|
|
}
|
|
var topResults []SearchResult
|
|
// Process vectors one by one to avoid loading everything into memory
|
|
for rows.Next() {
|
|
var (
|
|
embeddingsBlob []byte
|
|
slug, rawText, fileName string
|
|
)
|
|
|
|
if err := rows.Scan(&embeddingsBlob, &slug, &rawText, &fileName); err != nil {
|
|
vs.logger.Error("failed to scan row", "error", err)
|
|
continue
|
|
}
|
|
|
|
storedEmbeddings := DeserializeVector(embeddingsBlob)
|
|
|
|
// Calculate cosine similarity (returns value between -1 and 1, where 1 is most similar)
|
|
similarity := cosineSimilarity(query, storedEmbeddings)
|
|
distance := 1 - similarity // Convert to distance where 0 is most similar
|
|
|
|
result := SearchResult{
|
|
vector: models.VectorRow{
|
|
Embeddings: storedEmbeddings,
|
|
Slug: slug,
|
|
RawText: rawText,
|
|
FileName: fileName,
|
|
},
|
|
distance: distance,
|
|
}
|
|
|
|
// Add to top results and maintain only top 3
|
|
topResults = append(topResults, result)
|
|
|
|
// Sort and keep only top 3
|
|
sort.Slice(topResults, func(i, j int) bool {
|
|
return topResults[i].distance < topResults[j].distance
|
|
})
|
|
|
|
if len(topResults) > 3 {
|
|
topResults = topResults[:3] // Keep only closest 3
|
|
}
|
|
}
|
|
|
|
// Convert back to VectorRow slice
|
|
results := make([]models.VectorRow, 0, len(topResults))
|
|
for _, result := range topResults {
|
|
result.vector.Distance = result.distance
|
|
results = append(results, result.vector)
|
|
}
|
|
return results, nil
|
|
}
|
|
|
|
// ListFiles returns a list of all loaded files
|
|
func (vs *VectorStorage) ListFiles() ([]string, error) {
|
|
fileLists := make([][]string, 0)
|
|
// Query all supported tables and combine results
|
|
embeddingSizes := []int{384, 768, 1024, 1536, 2048, 3072, 4096, 5120}
|
|
for _, size := range embeddingSizes {
|
|
table := fmt.Sprintf("embeddings_%d", size)
|
|
query := "SELECT DISTINCT filename FROM " + table
|
|
rows, err := vs.sqlxDB.Query(query)
|
|
if err != nil {
|
|
// Continue if one table doesn't exist
|
|
continue
|
|
}
|
|
|
|
var files []string
|
|
for rows.Next() {
|
|
var filename string
|
|
if err := rows.Scan(&filename); err != nil {
|
|
continue
|
|
}
|
|
files = append(files, filename)
|
|
}
|
|
rows.Close()
|
|
|
|
fileLists = append(fileLists, files)
|
|
}
|
|
|
|
// Combine and deduplicate
|
|
fileSet := make(map[string]bool)
|
|
var allFiles []string
|
|
for _, files := range fileLists {
|
|
for _, file := range files {
|
|
if !fileSet[file] {
|
|
fileSet[file] = true
|
|
allFiles = append(allFiles, file)
|
|
}
|
|
}
|
|
}
|
|
return allFiles, nil
|
|
}
|
|
|
|
// RemoveEmbByFileName removes all embeddings associated with a specific filename
|
|
func (vs *VectorStorage) RemoveEmbByFileName(filename string) error {
|
|
var errors []string
|
|
embeddingSizes := []int{384, 768, 1024, 1536, 2048, 3072, 4096, 5120}
|
|
for _, size := range embeddingSizes {
|
|
table := fmt.Sprintf("embeddings_%d", size)
|
|
query := fmt.Sprintf("DELETE FROM %s WHERE filename = ?", table)
|
|
if _, err := vs.sqlxDB.Exec(query, filename); err != nil {
|
|
errors = append(errors, err.Error())
|
|
}
|
|
}
|
|
if len(errors) > 0 {
|
|
return fmt.Errorf("errors occurred: %s", strings.Join(errors, "; "))
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// cosineSimilarity calculates the cosine similarity between two vectors
|
|
func cosineSimilarity(a, b []float32) float32 {
|
|
if len(a) != len(b) {
|
|
return 0.0
|
|
}
|
|
var dotProduct, normA, normB float32
|
|
for i := 0; i < len(a); i++ {
|
|
dotProduct += a[i] * b[i]
|
|
normA += a[i] * a[i]
|
|
normB += b[i] * b[i]
|
|
}
|
|
if normA == 0 || normB == 0 {
|
|
return 0.0
|
|
}
|
|
return dotProduct / (sqrt(normA) * sqrt(normB))
|
|
}
|
|
|
|
// sqrt returns the square root of a float32
|
|
func sqrt(f float32) float32 {
|
|
// A simple implementation of square root using Newton's method
|
|
if f == 0 {
|
|
return 0
|
|
}
|
|
guess := f / 2
|
|
for i := 0; i < 10; i++ { // 10 iterations should be enough for good precision
|
|
guess = (guess + f/guess) / 2
|
|
}
|
|
return guess
|
|
}
|