最近想实现个人的应用调用gpt，并返回内容

一开始想尝试本地部署大模型，再使用对应API，在查看了各个模型需求的配置后

感觉自己的电脑有些吃力，就选择调用现有大模型的API来使用

因为个人的水平选择了百度千帆大模型平台（该平台提供了Nodejs版本的SDK）

Node.js SDK文档

@baiducloud/qianfan - npm (npmjs.com)

要实现原版gpt的打字机效果等，最好还是使用流式输出以及websocket

chat.gateway.ts

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
import { ChatCompletion, setEnvVariable } from "@baiducloud/qianfan";
import { ConnectedSocket, MessageBody, SubscribeMessage, WebSocketGateway, WebSocketServer } from '@nestjs/websockets';
import * as WebSocket from 'ws';

// 使用安全认证AK/SK鉴权，通过环境变量初始化；
setEnvVariable('QIANFAN_ACCESS_KEY', '自己的AK');
setEnvVariable('QIANFAN_SECRET_KEY', '自己的SK');

// websocket端口
@WebSocketGateway(4000)
export class ChatGateWay {
    private client = null

    constructor() {
        this.client = new ChatCompletion();
    }
    @SubscribeMessage('chat')
    // 订阅消息
    async chat(@MessageBody() data, @ConnectedSocket() client: WebSocket) {
        // 输入到千帆大模型
        const stream = await this.client.chat({
            messages: [
                {
                    role: 'user',
                    content: data,
                },
            ],
            stream: true
        }, 'ERNIE-3.5-8K');
        console.log(data)
        client.send(JSON.stringify({ event: 'start', data: 1 }));
        for await (const chunk of stream as AsyncIterableIterator<any>) {
            // 流式输出并发布消息
            client.send(JSON.stringify({ event: 'msg', data: JSON.stringify(chunk) }));
        }
        client.send(JSON.stringify({ event: 'end', data: 1 }));
    }
}

ws.adapter.ts

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
import * as WebSocket from 'ws';
import { WebSocketAdapter, INestApplicationContext } from '@nestjs/common';
import { MessageMappingProperties } from '@nestjs/websockets';
import { Observable, fromEvent, EMPTY } from 'rxjs';
import { mergeMap, filter } from 'rxjs/operators';

export class WsAdapter implements WebSocketAdapter {

    constructor(private app: INestApplicationContext) { }

    create(port: number, options: any = {}): any {
        console.log('ws create')
        return new WebSocket.Server({ port, ...options });
    }

    bindClientConnect(server, callback: Function) {
        // console.log('ws bindClientConnect, server:\n', server);
        server.on('connection', callback);
    }

    bindMessageHandlers(
        client: WebSocket,
        handlers: MessageMappingProperties[],
        process: (data: any) => Observable<any>,
    ) {
        // console.log('[waAdapter]有新的连接进来')
        fromEvent(client, 'message')
            .pipe(
                mergeMap(data => this.bindMessageHandler(client, data, handlers, process)),
                filter(result => result),
            )
            .subscribe(response => client.send(JSON.stringify(response)));
    }

    bindMessageHandler(
        client: WebSocket,
        buffer,
        handlers: MessageMappingProperties[],
        process: (data: any) => Observable<any>,
    ): Observable<any> {
        let message = null;
        console.log(buffer.data)
        try {
            message = JSON.parse(buffer.data);
        } catch (error) {
            console.log('ws解析json出错', error);
            return EMPTY;
        }

        const messageHandler = handlers.find(
            handler => handler.message === message.event,
        );
        if (!messageHandler) {
            return EMPTY;
        }
        return process(messageHandler.callback(message.data));
    }

    close(server) {
        console.log('ws server close');
        server.close();
    }
}

main.ts 使用适配器

1
app.useWebSocketAdapter(new WsAdapter(app));

效果图