-
-
Notifications
You must be signed in to change notification settings - Fork 1.4k
/
embeddings.go
267 lines (232 loc) · 9.56 KB
/
embeddings.go
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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
package openai
import (
"context"
"encoding/base64"
"encoding/binary"
"errors"
"math"
"net/http"
)
var ErrVectorLengthMismatch = errors.New("vector length mismatch")
// EmbeddingModel enumerates the models which can be used
// to generate Embedding vectors.
type EmbeddingModel string
const (
// Deprecated: The following block is shut down. Use text-embedding-ada-002 instead.
AdaSimilarity EmbeddingModel = "text-similarity-ada-001"
BabbageSimilarity EmbeddingModel = "text-similarity-babbage-001"
CurieSimilarity EmbeddingModel = "text-similarity-curie-001"
DavinciSimilarity EmbeddingModel = "text-similarity-davinci-001"
AdaSearchDocument EmbeddingModel = "text-search-ada-doc-001"
AdaSearchQuery EmbeddingModel = "text-search-ada-query-001"
BabbageSearchDocument EmbeddingModel = "text-search-babbage-doc-001"
BabbageSearchQuery EmbeddingModel = "text-search-babbage-query-001"
CurieSearchDocument EmbeddingModel = "text-search-curie-doc-001"
CurieSearchQuery EmbeddingModel = "text-search-curie-query-001"
DavinciSearchDocument EmbeddingModel = "text-search-davinci-doc-001"
DavinciSearchQuery EmbeddingModel = "text-search-davinci-query-001"
AdaCodeSearchCode EmbeddingModel = "code-search-ada-code-001"
AdaCodeSearchText EmbeddingModel = "code-search-ada-text-001"
BabbageCodeSearchCode EmbeddingModel = "code-search-babbage-code-001"
BabbageCodeSearchText EmbeddingModel = "code-search-babbage-text-001"
AdaEmbeddingV2 EmbeddingModel = "text-embedding-ada-002"
SmallEmbedding3 EmbeddingModel = "text-embedding-3-small"
LargeEmbedding3 EmbeddingModel = "text-embedding-3-large"
)
// Embedding is a special format of data representation that can be easily utilized by machine
// learning models and algorithms. The embedding is an information dense representation of the
// semantic meaning of a piece of text. Each embedding is a vector of floating point numbers,
// such that the distance between two embeddings in the vector space is correlated with semantic similarity
// between two inputs in the original format. For example, if two texts are similar,
// then their vector representations should also be similar.
type Embedding struct {
Object string `json:"object"`
Embedding []float32 `json:"embedding"`
Index int `json:"index"`
}
// DotProduct calculates the dot product of the embedding vector with another
// embedding vector. Both vectors must have the same length; otherwise, an
// ErrVectorLengthMismatch is returned. The method returns the calculated dot
// product as a float32 value.
func (e *Embedding) DotProduct(other *Embedding) (float32, error) {
if len(e.Embedding) != len(other.Embedding) {
return 0, ErrVectorLengthMismatch
}
var dotProduct float32
for i := range e.Embedding {
dotProduct += e.Embedding[i] * other.Embedding[i]
}
return dotProduct, nil
}
// EmbeddingResponse is the response from a Create embeddings request.
type EmbeddingResponse struct {
Object string `json:"object"`
Data []Embedding `json:"data"`
Model EmbeddingModel `json:"model"`
Usage Usage `json:"usage"`
httpHeader
}
type base64String string
func (b base64String) Decode() ([]float32, error) {
decodedData, err := base64.StdEncoding.DecodeString(string(b))
if err != nil {
return nil, err
}
const sizeOfFloat32 = 4
floats := make([]float32, len(decodedData)/sizeOfFloat32)
for i := 0; i < len(floats); i++ {
floats[i] = math.Float32frombits(binary.LittleEndian.Uint32(decodedData[i*4 : (i+1)*4]))
}
return floats, nil
}
// Base64Embedding is a container for base64 encoded embeddings.
type Base64Embedding struct {
Object string `json:"object"`
Embedding base64String `json:"embedding"`
Index int `json:"index"`
}
// EmbeddingResponseBase64 is the response from a Create embeddings request with base64 encoding format.
type EmbeddingResponseBase64 struct {
Object string `json:"object"`
Data []Base64Embedding `json:"data"`
Model EmbeddingModel `json:"model"`
Usage Usage `json:"usage"`
httpHeader
}
// ToEmbeddingResponse converts an embeddingResponseBase64 to an EmbeddingResponse.
func (r *EmbeddingResponseBase64) ToEmbeddingResponse() (EmbeddingResponse, error) {
data := make([]Embedding, len(r.Data))
for i, base64Embedding := range r.Data {
embedding, err := base64Embedding.Embedding.Decode()
if err != nil {
return EmbeddingResponse{}, err
}
data[i] = Embedding{
Object: base64Embedding.Object,
Embedding: embedding,
Index: base64Embedding.Index,
}
}
return EmbeddingResponse{
Object: r.Object,
Model: r.Model,
Data: data,
Usage: r.Usage,
}, nil
}
type EmbeddingRequestConverter interface {
// Needs to be of type EmbeddingRequestStrings or EmbeddingRequestTokens
Convert() EmbeddingRequest
}
// EmbeddingEncodingFormat is the format of the embeddings data.
// Currently, only "float" and "base64" are supported, however, "base64" is not officially documented.
// If not specified OpenAI will use "float".
type EmbeddingEncodingFormat string
const (
EmbeddingEncodingFormatFloat EmbeddingEncodingFormat = "float"
EmbeddingEncodingFormatBase64 EmbeddingEncodingFormat = "base64"
)
type EmbeddingRequest struct {
Input any `json:"input"`
Model EmbeddingModel `json:"model"`
User string `json:"user"`
EncodingFormat EmbeddingEncodingFormat `json:"encoding_format,omitempty"`
// Dimensions The number of dimensions the resulting output embeddings should have.
// Only supported in text-embedding-3 and later models.
Dimensions int `json:"dimensions,omitempty"`
}
func (r EmbeddingRequest) Convert() EmbeddingRequest {
return r
}
// EmbeddingRequestStrings is the input to a create embeddings request with a slice of strings.
type EmbeddingRequestStrings struct {
// Input is a slice of strings for which you want to generate an Embedding vector.
// Each input must not exceed 8192 tokens in length.
// OpenAPI suggests replacing newlines (\n) in your input with a single space, as they
// have observed inferior results when newlines are present.
// E.g.
// "The food was delicious and the waiter..."
Input []string `json:"input"`
// ID of the model to use. You can use the List models API to see all of your available models,
// or see our Model overview for descriptions of them.
Model EmbeddingModel `json:"model"`
// A unique identifier representing your end-user, which will help OpenAI to monitor and detect abuse.
User string `json:"user"`
// EmbeddingEncodingFormat is the format of the embeddings data.
// Currently, only "float" and "base64" are supported, however, "base64" is not officially documented.
// If not specified OpenAI will use "float".
EncodingFormat EmbeddingEncodingFormat `json:"encoding_format,omitempty"`
// Dimensions The number of dimensions the resulting output embeddings should have.
// Only supported in text-embedding-3 and later models.
Dimensions int `json:"dimensions,omitempty"`
}
func (r EmbeddingRequestStrings) Convert() EmbeddingRequest {
return EmbeddingRequest{
Input: r.Input,
Model: r.Model,
User: r.User,
EncodingFormat: r.EncodingFormat,
Dimensions: r.Dimensions,
}
}
type EmbeddingRequestTokens struct {
// Input is a slice of slices of ints ([][]int) for which you want to generate an Embedding vector.
// Each input must not exceed 8192 tokens in length.
// OpenAPI suggests replacing newlines (\n) in your input with a single space, as they
// have observed inferior results when newlines are present.
// E.g.
// "The food was delicious and the waiter..."
Input [][]int `json:"input"`
// ID of the model to use. You can use the List models API to see all of your available models,
// or see our Model overview for descriptions of them.
Model EmbeddingModel `json:"model"`
// A unique identifier representing your end-user, which will help OpenAI to monitor and detect abuse.
User string `json:"user"`
// EmbeddingEncodingFormat is the format of the embeddings data.
// Currently, only "float" and "base64" are supported, however, "base64" is not officially documented.
// If not specified OpenAI will use "float".
EncodingFormat EmbeddingEncodingFormat `json:"encoding_format,omitempty"`
// Dimensions The number of dimensions the resulting output embeddings should have.
// Only supported in text-embedding-3 and later models.
Dimensions int `json:"dimensions,omitempty"`
}
func (r EmbeddingRequestTokens) Convert() EmbeddingRequest {
return EmbeddingRequest{
Input: r.Input,
Model: r.Model,
User: r.User,
EncodingFormat: r.EncodingFormat,
Dimensions: r.Dimensions,
}
}
// CreateEmbeddings returns an EmbeddingResponse which will contain an Embedding for every item in |body.Input|.
// https://beta.openai.com/docs/api-reference/embeddings/create
//
// Body should be of type EmbeddingRequestStrings for embedding strings or EmbeddingRequestTokens
// for embedding groups of text already converted to tokens.
func (c *Client) CreateEmbeddings(
ctx context.Context,
conv EmbeddingRequestConverter,
) (res EmbeddingResponse, err error) {
baseReq := conv.Convert()
req, err := c.newRequest(
ctx,
http.MethodPost,
c.fullURL("/embeddings", withModel(string(baseReq.Model))),
withBody(baseReq),
)
if err != nil {
return
}
if baseReq.EncodingFormat != EmbeddingEncodingFormatBase64 {
err = c.sendRequest(req, &res)
return
}
base64Response := &EmbeddingResponseBase64{}
err = c.sendRequest(req, base64Response)
if err != nil {
return
}
res, err = base64Response.ToEmbeddingResponse()
return
}