1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
|
package storage
import (
"encoding/binary"
"fmt"
"gf-lt/models"
"sort"
"unsafe"
"github.com/jmoiron/sqlx"
)
type VectorRepo interface {
WriteVector(*models.VectorRow) error
SearchClosest(q []float32, limit int) ([]models.VectorRow, error)
ListFiles() ([]string, error)
RemoveEmbByFileName(filename string) error
DB() *sqlx.DB
}
// 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))
}
func fetchTableName(emb []float32) (string, error) {
switch len(emb) {
case 768:
return "embeddings_768", nil
default:
return "", fmt.Errorf("no table for the size of %d", len(emb))
}
}
func (p ProviderSQL) WriteVector(row *models.VectorRow) error {
tableName, err := fetchTableName(row.Embeddings)
if err != nil {
return err
}
serializedEmbeddings := SerializeVector(row.Embeddings)
query := fmt.Sprintf("INSERT INTO %s(embeddings, slug, raw_text, filename) VALUES (?, ?, ?, ?)", tableName)
_, err = p.db.Exec(query, serializedEmbeddings, row.Slug, row.RawText, row.FileName)
return err
}
func (p ProviderSQL) SearchClosest(q []float32, limit int) ([]models.VectorRow, error) {
tableName, err := fetchTableName(q)
if err != nil {
return nil, err
}
querySQL := "SELECT embeddings, slug, raw_text, filename FROM " + tableName
rows, err := p.db.Query(querySQL)
if err != nil {
return nil, err
}
defer rows.Close()
type SearchResult struct {
vector models.VectorRow
distance float32
}
var allResults []SearchResult
for rows.Next() {
var (
embeddingsBlob []byte
slug, rawText, fileName string
)
if err := rows.Scan(&embeddingsBlob, &slug, &rawText, &fileName); err != nil {
continue
}
storedEmbeddings := DeserializeVector(embeddingsBlob)
// Calculate cosine similarity (returns value between -1 and 1, where 1 is most similar)
similarity := cosineSimilarity(q, 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,
}
allResults = append(allResults, result)
}
// Sort by distance
sort.Slice(allResults, func(i, j int) bool {
return allResults[i].distance < allResults[j].distance
})
// Truncate to limit
if len(allResults) > limit {
allResults = allResults[:limit]
}
// Convert back to VectorRow slice
results := make([]models.VectorRow, len(allResults))
for i, result := range allResults {
result.vector.Distance = result.distance
results[i] = result.vector
}
return results, 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
}
func (p ProviderSQL) ListFiles() ([]string, error) {
query := "SELECT DISTINCT filename FROM embeddings_768"
rows, err := p.db.Query(query)
if err != nil {
return nil, err
}
defer rows.Close()
var allFiles []string
for rows.Next() {
var filename string
if err := rows.Scan(&filename); err != nil {
continue
}
allFiles = append(allFiles, filename)
}
return allFiles, nil
}
func (p ProviderSQL) RemoveEmbByFileName(filename string) error {
query := "DELETE FROM embeddings_768 WHERE filename = ?"
_, err := p.db.Exec(query, filename)
return err
}
|
