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
|
import {
GenerativeModel,
GoogleGenerativeAI,
type Content,
type GenerateContentResult,
} from "@google/generative-ai";
import { RESPONSE_LENGTH } from "./logic/constants";
import type { AResult, ChatMessage, OChoice, OChunk, OMessage } from "./types";
export default class Conversation {
private tokenizer: (text: string) => number;
private maxTokens: number;
private model: GenerativeModel;
constructor(
maxTokens = 200_000,
tokenizer: (text: string) => number = (text) => text.length / 3,
) {
this.maxTokens = maxTokens;
this.tokenizer = tokenizer;
const gem = new GoogleGenerativeAI(Bun.env["GEMINI_API_KEY"]!);
this.model = gem.getGenerativeModel({
model: "gemini-2.0-flash-exp",
generationConfig: { maxOutputTokens: RESPONSE_LENGTH },
});
}
public setModel(model: string) {
const gem = new GoogleGenerativeAI(Bun.env["GEMINI_API_KEY"]!);
this.model = gem.getGenerativeModel({
model,
generationConfig: { maxOutputTokens: RESPONSE_LENGTH },
});
}
private mapMessages(input: ChatMessage[]): Content[] {
return input.map((m) => ({
role: m.author === "gemini" ? "model" : "user",
parts: [{ text: m.text }],
}));
}
private mapMessagesR1(input: ChatMessage[]): Content[] {
return input.reduce((acc: Content[], m, i) => {
const prev = acc[i - 1];
const role = m.author === "gemini" ? "model" : "user";
const msg = { role, parts: [{ text: m.text }] };
if (prev?.role === role) acc[i - 1] = msg;
else acc = [...acc, msg];
return acc;
}, []);
}
private async apiCall(
messages: Content[],
isR1: boolean = false,
): Promise<AResult<string[]>> {
try {
const chat = this.model.startChat({ history: messages });
const res = await chat.sendMessage("");
return { ok: [res.response.text()] };
} catch (e) {
console.log(e, "error in gemini api");
return { error: `${e}` };
}
}
private async apiCallStream(
messages: Content[],
handle: (c: any) => void,
isR1: boolean = false,
): Promise<void> {
try {
const chat = this.model.startChat({ history: messages });
const res = await chat.sendMessage("");
// for await (const chunk of res.stream()) {
// handle(chunk.text());
// }
} catch (e) {
console.log(e, "error in gemini api");
handle(`Error streaming Gemini, ${e}`);
}
}
public async send(sys: string, input: ChatMessage[]) {
const messages = this.mapMessages(input);
const truncated = this.truncateHistory(messages);
const res = await this.apiCall(truncated);
return res;
}
public async sendR1(input: ChatMessage[]) {
const messages = this.mapMessagesR1(input);
const truncated = this.truncateHistory(messages);
const res = await this.apiCall(truncated, true);
return res;
}
public async stream(
sys: string,
input: ChatMessage[],
handle: (c: any) => void,
) {
const messages = this.mapMessages(input);
const truncated = this.truncateHistory(messages);
await this.apiCallStream(truncated, handle);
}
public async streamR1(input: ChatMessage[], handle: (c: any) => void) {
const messages = this.mapMessagesR1(input);
const truncated = this.truncateHistory(messages);
await this.apiCallStream(truncated, handle, true);
}
public async sendDoc(data: ArrayBuffer, mimeType: string, prompt: string) {
const res = await this.model.generateContent([
{
inlineData: {
data: Buffer.from(data).toString("base64"),
mimeType,
},
},
prompt,
]);
return res;
}
private truncateHistory(messages: Content[]): Content[] {
const totalTokens = messages.reduce((total, message) => {
return total + this.tokenizer(message.parts[0].text || "");
}, 0);
while (totalTokens > this.maxTokens && messages.length > 1) {
messages.splice(0, 1);
}
return messages;
}
}
|
