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
|
package main
import (
"bufio"
"bytes"
"elefant/models"
"elefant/storage"
"encoding/json"
"fmt"
"io"
"log/slog"
"net/http"
"os"
"strings"
"time"
"github.com/rivo/tview"
)
var httpClient = http.Client{
Timeout: time.Second * 20,
}
var (
logger *slog.Logger
APIURL = "http://localhost:8080/v1/chat/completions"
userRole = "user"
assistantRole = "assistant"
toolRole = "tool"
assistantIcon = "<🤖>: "
userIcon = "<user>: "
// TODO: pass as an cli arg or have config
logFileName = "log.txt"
showSystemMsgs bool
chunkLimit = 1000
activeChatName string
chunkChan = make(chan string, 10)
streamDone = make(chan bool, 1)
chatBody *models.ChatBody
store storage.FullRepo
defaultFirstMsg = "Hello! What can I do for you?"
defaultStarter = []models.MessagesStory{
{Role: "system", Content: systemMsg},
{Role: assistantRole, Content: defaultFirstMsg},
}
defaultStarterBytes, _ = json.Marshal(chatBody.Messages)
interruptResp = false
)
// ====
func formMsg(chatBody *models.ChatBody, newMsg, role string) io.Reader {
if newMsg != "" { // otherwise let the bot continue
newMsg := models.MessagesStory{Role: role, Content: newMsg}
chatBody.Messages = append(chatBody.Messages, newMsg)
}
data, err := json.Marshal(chatBody)
if err != nil {
logger.Error("failed to form a msg", "error", err)
return nil
}
return bytes.NewReader(data)
}
// func sendMsgToLLM(body io.Reader) (*models.LLMRespChunk, error) {
func sendMsgToLLM(body io.Reader) (any, error) {
resp, err := httpClient.Post(APIURL, "application/json", body)
if err != nil {
logger.Error("llamacpp api", "error", err)
return nil, err
}
defer resp.Body.Close()
llmResp := []models.LLMRespChunk{}
// chunkChan <- assistantIcon
reader := bufio.NewReader(resp.Body)
counter := 0
for {
if interruptResp {
interruptResp = false
logger.Info("interrupted bot response")
break
}
llmchunk := models.LLMRespChunk{}
if counter > chunkLimit {
logger.Warn("response hit chunk limit", "limit", chunkLimit)
streamDone <- true
break
}
line, err := reader.ReadBytes('\n')
if err != nil {
streamDone <- true
logger.Error("error reading response body", "error", err)
}
// logger.Info("linecheck", "line", string(line), "len", len(line), "counter", counter)
if len(line) <= 1 {
continue // skip \n
}
// starts with -> data:
line = line[6:]
if err := json.Unmarshal(line, &llmchunk); err != nil {
logger.Error("failed to decode", "error", err, "line", string(line))
streamDone <- true
return nil, err
}
llmResp = append(llmResp, llmchunk)
// logger.Info("streamview", "chunk", llmchunk)
// if llmchunk.Choices[len(llmchunk.Choices)-1].FinishReason != "chat.completion.chunk" {
if llmchunk.Choices[len(llmchunk.Choices)-1].FinishReason == "stop" {
streamDone <- true
// last chunk
break
}
counter++
// bot sends way too many \n
answerText := strings.ReplaceAll(llmchunk.Choices[0].Delta.Content, "\n\n", "\n")
chunkChan <- answerText
}
return llmResp, nil
}
func chatRound(userMsg, role string, tv *tview.TextView) {
botRespMode = true
reader := formMsg(chatBody, userMsg, role)
if reader == nil {
return // any notification in that case?
}
go sendMsgToLLM(reader)
fmt.Fprintf(tv, fmt.Sprintf("(%d) ", len(chatBody.Messages)))
fmt.Fprintf(tv, assistantIcon)
respText := strings.Builder{}
out:
for {
select {
case chunk := <-chunkChan:
// fmt.Printf(chunk)
fmt.Fprintf(tv, chunk)
respText.WriteString(chunk)
tv.ScrollToEnd()
case <-streamDone:
break out
}
}
botRespMode = false
chatBody.Messages = append(chatBody.Messages, models.MessagesStory{
Role: assistantRole, Content: respText.String(),
})
// bot msg is done;
// now check it for func call
// logChat(activeChatName, chatBody.Messages)
err := updateStorageChat(activeChatName, chatBody.Messages)
if err != nil {
logger.Warn("failed to update storage", "error", err, "name", activeChatName)
}
findCall(respText.String(), tv)
}
func findCall(msg string, tv *tview.TextView) {
// prefix := "__tool_call__\n"
// suffix := "\n__tool_call__"
// if !strings.HasPrefix(msg, prefix) ||
// !strings.HasSuffix(msg, suffix) {
// return
// }
// jsStr := strings.TrimSuffix(strings.TrimPrefix(msg, prefix), suffix)
fc := models.FuncCall{}
jsStr := toolCallRE.FindString(msg)
if jsStr == "" {
// tool call not found
return
}
if err := json.Unmarshal([]byte(jsStr), &fc); err != nil {
logger.Error("failed to unmarshal tool call", "error", err)
return
}
// call a func
f, ok := fnMap[fc.Name]
if !ok {
m := fmt.Sprintf("%s is not implemented", fc.Name)
chatRound(m, toolRole, tv)
return
}
resp := f(fc.Args)
toolMsg := fmt.Sprintf("tool response: %+v", resp)
// reader := formMsg(chatBody, toolMsg, toolRole)
// sendMsgToLLM()
chatRound(toolMsg, toolRole, tv)
// return func result to the llm
}
func chatToTextSlice(showSys bool) []string {
resp := make([]string, len(chatBody.Messages))
for i, msg := range chatBody.Messages {
if !showSys && (msg.Role != assistantRole && msg.Role != userRole) {
continue
}
resp[i] = msg.ToText(i)
}
return resp
}
func chatToText(showSys bool) string {
s := chatToTextSlice(showSys)
return strings.Join(s, "")
}
func textToMsg(rawMsg string) models.MessagesStory {
msg := models.MessagesStory{}
// system and tool?
if strings.HasPrefix(rawMsg, assistantIcon) {
msg.Role = assistantRole
msg.Content = strings.TrimPrefix(rawMsg, assistantIcon)
return msg
}
if strings.HasPrefix(rawMsg, userIcon) {
msg.Role = userRole
msg.Content = strings.TrimPrefix(rawMsg, userIcon)
return msg
}
return msg
}
func textSliceToChat(chat []string) []models.MessagesStory {
resp := make([]models.MessagesStory, len(chat))
for i, rawMsg := range chat {
msg := textToMsg(rawMsg)
resp[i] = msg
}
return resp
}
func init() {
file, err := os.OpenFile(logFileName, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0644)
if err != nil {
logger.Error("failed to open log file", "error", err, "filename", logFileName)
return
}
logger = slog.New(slog.NewTextHandler(file, nil))
store = storage.NewProviderSQL("test.db", logger)
// https://github.com/coreydaley/ggerganov-llama.cpp/blob/master/examples/server/README.md
// load all chats in memory
loadHistoryChats()
lastChat := loadOldChatOrGetNew()
logger.Info("loaded history", "chat", lastChat)
chatBody = &models.ChatBody{
Model: "modl_name",
Stream: true,
Messages: lastChat,
}
}
|