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
|
package main
import (
"bytes"
"elefant/models"
"encoding/json"
"io"
"strings"
)
type ChunkParser interface {
ParseChunk([]byte) (string, bool, error)
FormMsg(msg, role string, cont bool) (io.Reader, error)
}
func choseChunkParser() {
chunkParser = LlamaCPPeer{}
switch cfg.CurrentAPI {
case "http://localhost:8080/completion":
chunkParser = LlamaCPPeer{}
case "http://localhost:8080/v1/chat/completions":
chunkParser = OpenAIer{}
case "https://api.deepseek.com/beta/completions":
chunkParser = DeepSeeker{}
default:
chunkParser = LlamaCPPeer{}
}
// if strings.Contains(cfg.CurrentAPI, "chat") {
// logger.Debug("chosen chat parser")
// chunkParser = OpenAIer{}
// return
// }
// logger.Debug("chosen llamacpp /completion parser")
}
type LlamaCPPeer struct {
}
type OpenAIer struct {
}
type DeepSeeker struct {
}
func (lcp LlamaCPPeer) FormMsg(msg, role string, resume bool) (io.Reader, error) {
if msg != "" { // otherwise let the bot to continue
newMsg := models.RoleMsg{Role: role, Content: msg}
chatBody.Messages = append(chatBody.Messages, newMsg)
// if rag
if cfg.RAGEnabled {
ragResp, err := chatRagUse(newMsg.Content)
if err != nil {
logger.Error("failed to form a rag msg", "error", err)
return nil, err
}
ragMsg := models.RoleMsg{Role: cfg.ToolRole, Content: ragResp}
chatBody.Messages = append(chatBody.Messages, ragMsg)
}
}
if cfg.ToolUse && !resume {
// add to chat body
chatBody.Messages = append(chatBody.Messages, models.RoleMsg{Role: cfg.ToolRole, Content: toolSysMsg})
}
messages := make([]string, len(chatBody.Messages))
for i, m := range chatBody.Messages {
messages[i] = m.ToPrompt()
}
prompt := strings.Join(messages, "\n")
// strings builder?
if !resume {
botMsgStart := "\n" + cfg.AssistantRole + ":\n"
prompt += botMsgStart
}
if cfg.ThinkUse && !cfg.ToolUse {
prompt += "<think>"
}
logger.Debug("checking prompt for /completion", "tool_use", cfg.ToolUse,
"msg", msg, "resume", resume, "prompt", prompt)
var payload any
payload = models.NewLCPReq(prompt, cfg, defaultLCPProps)
if strings.Contains(chatBody.Model, "deepseek") {
payload = models.NewDSCompletionReq(prompt, chatBody.Model,
defaultLCPProps["temp"], cfg)
}
data, err := json.Marshal(payload)
if err != nil {
logger.Error("failed to form a msg", "error", err)
return nil, err
}
return bytes.NewReader(data), nil
}
func (lcp LlamaCPPeer) ParseChunk(data []byte) (string, bool, error) {
llmchunk := models.LlamaCPPResp{}
if err := json.Unmarshal(data, &llmchunk); err != nil {
logger.Error("failed to decode", "error", err, "line", string(data))
return "", false, err
}
if llmchunk.Stop {
if llmchunk.Content != "" {
logger.Error("text inside of finish llmchunk", "chunk", llmchunk)
}
return llmchunk.Content, true, nil
}
return llmchunk.Content, false, nil
}
func (op OpenAIer) ParseChunk(data []byte) (string, bool, error) {
llmchunk := models.LLMRespChunk{}
if err := json.Unmarshal(data, &llmchunk); err != nil {
logger.Error("failed to decode", "error", err, "line", string(data))
return "", false, err
}
content := llmchunk.Choices[len(llmchunk.Choices)-1].Delta.Content
if llmchunk.Choices[len(llmchunk.Choices)-1].FinishReason == "stop" {
if content != "" {
logger.Error("text inside of finish llmchunk", "chunk", llmchunk)
}
return content, true, nil
}
return content, false, nil
}
func (op OpenAIer) FormMsg(msg, role string, resume bool) (io.Reader, error) {
if cfg.ToolUse && !resume {
// prompt += "\n" + cfg.ToolRole + ":\n" + toolSysMsg
// add to chat body
chatBody.Messages = append(chatBody.Messages, models.RoleMsg{Role: cfg.ToolRole, Content: toolSysMsg})
}
if msg != "" { // otherwise let the bot continue
newMsg := models.RoleMsg{Role: role, Content: msg}
chatBody.Messages = append(chatBody.Messages, newMsg)
// if rag
if cfg.RAGEnabled {
ragResp, err := chatRagUse(newMsg.Content)
if err != nil {
logger.Error("failed to form a rag msg", "error", err)
return nil, err
}
ragMsg := models.RoleMsg{Role: cfg.ToolRole, Content: ragResp}
chatBody.Messages = append(chatBody.Messages, ragMsg)
}
}
data, err := json.Marshal(chatBody)
if err != nil {
logger.Error("failed to form a msg", "error", err)
return nil, err
}
return bytes.NewReader(data), nil
}
// deepseek
func (ds DeepSeeker) ParseChunk(data []byte) (string, bool, error) {
llmchunk := models.DSCompletionResp{}
if err := json.Unmarshal(data, &llmchunk); err != nil {
logger.Error("failed to decode", "error", err, "line", string(data))
return "", false, err
}
if llmchunk.Choices[0].FinishReason != "" {
if llmchunk.Choices[0].Text != "" {
logger.Error("text inside of finish llmchunk", "chunk", llmchunk)
}
return llmchunk.Choices[0].Text, true, nil
}
return llmchunk.Choices[0].Text, false, nil
}
func (ds DeepSeeker) FormMsg(msg, role string, resume bool) (io.Reader, error) {
if msg != "" { // otherwise let the bot to continue
newMsg := models.RoleMsg{Role: role, Content: msg}
chatBody.Messages = append(chatBody.Messages, newMsg)
// if rag
if cfg.RAGEnabled {
ragResp, err := chatRagUse(newMsg.Content)
if err != nil {
logger.Error("failed to form a rag msg", "error", err)
return nil, err
}
ragMsg := models.RoleMsg{Role: cfg.ToolRole, Content: ragResp}
chatBody.Messages = append(chatBody.Messages, ragMsg)
}
}
if cfg.ToolUse && !resume {
// add to chat body
chatBody.Messages = append(chatBody.Messages, models.RoleMsg{Role: cfg.ToolRole, Content: toolSysMsg})
}
messages := make([]string, len(chatBody.Messages))
for i, m := range chatBody.Messages {
messages[i] = m.ToPrompt()
}
prompt := strings.Join(messages, "\n")
// strings builder?
if !resume {
botMsgStart := "\n" + cfg.AssistantRole + ":\n"
prompt += botMsgStart
}
if cfg.ThinkUse && !cfg.ToolUse {
prompt += "<think>"
}
logger.Debug("checking prompt for /completion", "tool_use", cfg.ToolUse,
"msg", msg, "resume", resume, "prompt", prompt)
payload := models.NewDSCompletionReq(prompt, chatBody.Model,
defaultLCPProps["temp"], cfg)
data, err := json.Marshal(payload)
if err != nil {
logger.Error("failed to form a msg", "error", err)
return nil, err
}
return bytes.NewReader(data), nil
}
|
