forked from pkoukk/tiktoken-go
-
Notifications
You must be signed in to change notification settings - Fork 8
/
encoding.go
235 lines (215 loc) · 7 KB
/
encoding.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
package tiktoken
import (
"errors"
"strings"
"sync"
)
const ENDOFTEXT string = "<|endoftext|>"
const FIM_PREFIX string = "<|fim_prefix|>"
const FIM_MIDDLE string = "<|fim_middle|>"
const FIM_SUFFIX string = "<|fim_suffix|>"
const ENDOFPROMPT string = "<|endofprompt|>"
const (
MODEL_O200K_BASE string = "o200k_base"
MODEL_CL100K_BASE string = "cl100k_base"
MODEL_P50K_BASE string = "p50k_base"
MODEL_P50K_EDIT string = "p50k_edit"
MODEL_R50K_BASE string = "r50k_base"
MODEL_GPT2 string = "gpt2"
)
var MODEL_TO_ENCODING = map[string]string{
// chat
"gpt-4o": MODEL_O200K_BASE,
"gpt-4": MODEL_CL100K_BASE,
"gpt-3.5-turbo": MODEL_CL100K_BASE,
"gpt-3.5": MODEL_CL100K_BASE, // Common shorthand
"gpt-35-turbo": MODEL_CL100K_BASE, // Azure deployment name
// base
"davinci-002": MODEL_CL100K_BASE,
"babbage-002": MODEL_CL100K_BASE,
// embeddings
"text-embedding-ada-002": MODEL_CL100K_BASE,
"text-embedding-3-small": MODEL_CL100K_BASE,
"text-embedding-3-large": MODEL_CL100K_BASE,
// DEPRECATED MODELS
// text (DEPRECATED)
"text-davinci-003": MODEL_P50K_BASE,
"text-davinci-002": MODEL_P50K_BASE,
"text-davinci-001": MODEL_P50K_BASE,
"text-curie-001": MODEL_P50K_BASE,
"text-babbage-001": MODEL_P50K_BASE,
"text-ada-001": MODEL_P50K_BASE,
"davinci": MODEL_P50K_BASE,
"curie": MODEL_P50K_BASE,
"babbage": MODEL_P50K_BASE,
"ada": MODEL_P50K_BASE,
// code (DEPRECATED)
"code-davinci-002": MODEL_P50K_BASE,
"code-davinci-001": MODEL_P50K_BASE,
"code-cushman-002": MODEL_P50K_BASE,
"code-cushman-001": MODEL_P50K_BASE,
"davinci-codex": MODEL_P50K_BASE,
"cushman-codex": MODEL_P50K_BASE,
// edit (DEPRECATED)
"text-davinci-edit-001": MODEL_P50K_EDIT,
"code-davinci-edit-001": MODEL_P50K_EDIT,
// old embeddings (DEPRECATED)
"text-similarity-davinci-001": MODEL_R50K_BASE,
"text-similarity-curie-001": MODEL_R50K_BASE,
"text-similarity-babbage-001": MODEL_R50K_BASE,
"text-similarity-ada-001": MODEL_R50K_BASE,
"text-search-davinci-doc-001": MODEL_R50K_BASE,
"text-search-curie-doc-001": MODEL_R50K_BASE,
"text-search-babbage-doc-001": MODEL_R50K_BASE,
"text-search-ada-doc-001": MODEL_R50K_BASE,
"code-search-babbage-code-001": MODEL_R50K_BASE,
"code-search-ada-code-001": MODEL_R50K_BASE,
// open source
"gpt2": MODEL_GPT2,
"gpt-2": MODEL_GPT2, // Maintains consistency with gpt-4
}
var MODEL_PREFIX_TO_ENCODING = map[string]string{
// chat
"gpt-4o-": MODEL_O200K_BASE, // e.g., gpt-4o-2024-05-13
"gpt-4-": MODEL_CL100K_BASE, // e.g., gpt-4-0314, etc., plus gpt-4-32k
"gpt-3.5-turbo-": MODEL_CL100K_BASE, // e.g, gpt-3.5-turbo-0301, -0401, etc.
"gpt-35-turbo-": MODEL_CL100K_BASE, // Azure deployment name
// fine-tuned
"ft:gpt-4": MODEL_CL100K_BASE,
"ft:gpt-3.5-turbo": MODEL_CL100K_BASE,
"ft:davinci-002": MODEL_CL100K_BASE,
"ft:babbage-002": MODEL_CL100K_BASE,
}
var encodingMap map[string]*Encoding
var l *sync.Mutex
func init() {
encodingMap = make(map[string]*Encoding)
l = &sync.Mutex{}
}
type Encoding struct {
Name string
PatStr string
MergeableRanks map[string]int
SpecialTokens map[string]int
ExplicitNVocab int
}
func getEncoding(encodingName string) (*Encoding, error) {
l.Lock()
defer l.Unlock()
if encoding, ok := encodingMap[encodingName]; ok {
return encoding, nil
}
initEncoding, err := initEncoding(encodingName)
if err != nil {
return nil, err
}
encodingMap[encodingName] = initEncoding
return encodingMap[encodingName], nil
}
func initEncoding(encodingName string) (*Encoding, error) {
switch encodingName {
case MODEL_O200K_BASE:
return o200k_base()
case MODEL_CL100K_BASE:
return cl100k_base()
case MODEL_P50K_BASE:
return p50k_base()
case MODEL_R50K_BASE:
return r50k_base()
case MODEL_P50K_EDIT:
return p50k_edit()
default:
return nil, errors.New("Unknown encoding: " + encodingName)
}
}
func o200k_base() (*Encoding, error) {
ranks, err := bpeLoader.LoadTiktokenBpe("https://openaipublic.blob.core.windows.net/encodings/o200k_base.tiktoken")
if err != nil {
return nil, err
}
special_tokens := map[string]int{
ENDOFTEXT: 199999,
ENDOFPROMPT: 200018,
}
patStr := []string{
`[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]*[\p{Ll}\p{Lm}\p{Lo}\p{M}]+(?i:'s|'t|'re|'ve|'m|'ll|'d)?`,
`[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]+[\p{Ll}\p{Lm}\p{Lo}\p{M}]*(?i:'s|'t|'re|'ve|'m|'ll|'d)?`,
`\p{N}{1,3}`,
` ?[^\s\p{L}\p{N}]+[\r\n/]*`,
`\s*[\r\n]+`,
`\s+(?!\S)`,
`\s+`,
}
return &Encoding{
Name: MODEL_O200K_BASE,
PatStr: strings.Join(patStr, "|"),
MergeableRanks: ranks,
SpecialTokens: special_tokens,
}, nil
}
func cl100k_base() (*Encoding, error) {
ranks, err := bpeLoader.LoadTiktokenBpe("https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken")
if err != nil {
return nil, err
}
special_tokens := map[string]int{
ENDOFTEXT: 100257,
FIM_PREFIX: 100258,
FIM_MIDDLE: 100259,
FIM_SUFFIX: 100260,
ENDOFPROMPT: 100276,
}
return &Encoding{
Name: MODEL_CL100K_BASE,
PatStr: `(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}{1,3}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
MergeableRanks: ranks,
SpecialTokens: special_tokens,
}, nil
}
func p50k_edit() (*Encoding, error) {
ranks, err := bpeLoader.LoadTiktokenBpe("https://openaipublic.blob.core.windows.net/encodings/p50k_base.tiktoken")
if err != nil {
return nil, err
}
special_tokens := map[string]int{ENDOFTEXT: 50256, FIM_PREFIX: 50281, FIM_MIDDLE: 50282, FIM_SUFFIX: 50283}
return &Encoding{
Name: MODEL_P50K_EDIT,
PatStr: `'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+`,
MergeableRanks: ranks,
SpecialTokens: special_tokens,
}, nil
}
func p50k_base() (*Encoding, error) {
ranks, err := bpeLoader.LoadTiktokenBpe("https://openaipublic.blob.core.windows.net/encodings/p50k_base.tiktoken")
if err != nil {
return nil, err
}
special_tokens := map[string]int{ENDOFTEXT: 50256}
// ExplicitNVocab := 50281
// max_tokens := int(math.Max(float64(len(special_tokens)), float64(len(ranks))))
// if len(special_tokens)+len(ranks) != max_tokens {
// return nil, errors.New("special_tokens and ranks must be disjoint")
// }
return &Encoding{
Name: MODEL_P50K_BASE,
PatStr: `'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+`,
MergeableRanks: ranks,
SpecialTokens: special_tokens,
ExplicitNVocab: 50281,
}, nil
}
func r50k_base() (*Encoding, error) {
ranks, err := bpeLoader.LoadTiktokenBpe("https://openaipublic.blob.core.windows.net/encodings/r50k_base.tiktoken")
if err != nil {
return nil, err
}
special_tokens := map[string]int{ENDOFTEXT: 50256}
return &Encoding{
Name: MODEL_R50K_BASE,
MergeableRanks: ranks,
PatStr: `'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+`,
SpecialTokens: special_tokens,
ExplicitNVocab: 50257,
}, nil
}
// var ENCODING_MAP = map[string]*Encoding{}