Delete index.js
Browse files
index.js
DELETED
|
@@ -1,268 +0,0 @@
|
|
| 1 |
-
import grpc from '@grpc/grpc-js';
|
| 2 |
-
import protoLoader from '@grpc/proto-loader';
|
| 3 |
-
import { AutoRouter, cors, error, json } from 'itty-router';
|
| 4 |
-
import dotenv from 'dotenv';
|
| 5 |
-
import { createServerAdapter } from '@whatwg-node/server';
|
| 6 |
-
import { createServer } from 'http';
|
| 7 |
-
|
| 8 |
-
// 加载环境变量
|
| 9 |
-
dotenv.config();
|
| 10 |
-
// 初始化配置
|
| 11 |
-
class Config {
|
| 12 |
-
constructor() {
|
| 13 |
-
this.API_PREFIX = process.env.API_PREFIX || '/';
|
| 14 |
-
this.API_KEY = process.env.API_KEY || '';
|
| 15 |
-
this.MAX_RETRY_COUNT = process.env.MAX_RETRY_COUNT || 3;
|
| 16 |
-
this.RETRY_DELAY = process.env.RETRY_DELAY || 5000;
|
| 17 |
-
this.COMMON_GRPC = 'runtime-native-io-vertex-inference-grpc-service-lmuw6mcn3q-ul.a.run.app';
|
| 18 |
-
this.COMMON_PROTO = './VertexInferenceService.proto';
|
| 19 |
-
this.GPT_GRPC = 'runtime-native-io-gpt-inference-grpc-service-lmuw6mcn3q-ul.a.run.app';
|
| 20 |
-
this.GPT_PROTO = './GPTInferenceService.proto';
|
| 21 |
-
this.PORT = process.env.PORT || 8787;
|
| 22 |
-
}
|
| 23 |
-
}
|
| 24 |
-
class GRPCHandler {
|
| 25 |
-
constructor(protoFilePath) {
|
| 26 |
-
// 动态加载传入的 .proto 文件路径
|
| 27 |
-
this.packageDefinition = protoLoader.loadSync(protoFilePath, {
|
| 28 |
-
keepCase: true,
|
| 29 |
-
longs: String,
|
| 30 |
-
enums: String,
|
| 31 |
-
defaults: true,
|
| 32 |
-
oneofs: true
|
| 33 |
-
});
|
| 34 |
-
}
|
| 35 |
-
}
|
| 36 |
-
const config = new Config();
|
| 37 |
-
// 中间件
|
| 38 |
-
// 添加运行回源
|
| 39 |
-
const { preflight, corsify } = cors({
|
| 40 |
-
origin: '*',
|
| 41 |
-
allowMethods: '*',
|
| 42 |
-
exposeHeaders: '*',
|
| 43 |
-
});
|
| 44 |
-
|
| 45 |
-
// 添加认证
|
| 46 |
-
const withAuth = (request) => {
|
| 47 |
-
if (config.API_KEY) {
|
| 48 |
-
const authHeader = request.headers.get('Authorization');
|
| 49 |
-
if (!authHeader || !authHeader.startsWith('Bearer ')) {
|
| 50 |
-
return error(401, 'Unauthorized: Missing or invalid Authorization header');
|
| 51 |
-
}
|
| 52 |
-
const token = authHeader.substring(7);
|
| 53 |
-
if (token !== config.API_KEY) {
|
| 54 |
-
return error(403, 'Forbidden: Invalid API key');
|
| 55 |
-
}
|
| 56 |
-
}
|
| 57 |
-
};
|
| 58 |
-
// 返回运行信息
|
| 59 |
-
const logger = (res, req) => {
|
| 60 |
-
console.log(req.method, res.status, req.url, Date.now() - req.start, 'ms');
|
| 61 |
-
};
|
| 62 |
-
const router = AutoRouter({
|
| 63 |
-
before: [preflight, withAuth],
|
| 64 |
-
missing: () => error(404, '404 not found.'),
|
| 65 |
-
finally: [corsify, logger],
|
| 66 |
-
});
|
| 67 |
-
// Router路径
|
| 68 |
-
router.get('/', () => json({ message: 'API 服务运行中~' }));
|
| 69 |
-
router.get('/ping', () => json({ message: 'pong' }));
|
| 70 |
-
router.post(config.API_PREFIX + '/v1/chat/completions', (req) => handleCompletion(req));
|
| 71 |
-
|
| 72 |
-
async function GrpcToPieces(models, message, rules, stream, temperature, top_p) {
|
| 73 |
-
// 在非GPT类型的模型中,temperature和top_p是无效的
|
| 74 |
-
// 使用系统的根证书
|
| 75 |
-
const credentials = grpc.credentials.createSsl();
|
| 76 |
-
let client, request;
|
| 77 |
-
if (models.includes('gpt')) {
|
| 78 |
-
// 加载proto文件
|
| 79 |
-
const packageDefinition = new GRPCHandler(config.GPT_PROTO).packageDefinition;
|
| 80 |
-
// 构建请求消息
|
| 81 |
-
request = {
|
| 82 |
-
models: models,
|
| 83 |
-
messages: [
|
| 84 |
-
{ role: 0, message: rules }, // system
|
| 85 |
-
{ role: 1, message: message } // user
|
| 86 |
-
],
|
| 87 |
-
temperature: temperature || 0.1,
|
| 88 |
-
top_p: top_p ?? 1,
|
| 89 |
-
}
|
| 90 |
-
// 获取gRPC对象
|
| 91 |
-
const GRPCobjects = grpc.loadPackageDefinition(packageDefinition).runtime.aot.machine_learning.parents.gpt;
|
| 92 |
-
client = new GRPCobjects.GPTInferenceService(config.GPT_GRPC, credentials);
|
| 93 |
-
} else {
|
| 94 |
-
// 加载proto文件
|
| 95 |
-
const packageDefinition = new GRPCHandler(config.COMMON_PROTO).packageDefinition;
|
| 96 |
-
// 构建请求消息
|
| 97 |
-
request = {
|
| 98 |
-
models: models,
|
| 99 |
-
args: {
|
| 100 |
-
messages: {
|
| 101 |
-
unknown: 1,
|
| 102 |
-
message: message
|
| 103 |
-
},
|
| 104 |
-
rules: rules
|
| 105 |
-
}
|
| 106 |
-
};
|
| 107 |
-
// 获取gRPC对象
|
| 108 |
-
const GRPCobjects = grpc.loadPackageDefinition(packageDefinition).runtime.aot.machine_learning.parents.vertex;
|
| 109 |
-
client = new GRPCobjects.VertexInferenceService(config.COMMON_GRPC, credentials);
|
| 110 |
-
}
|
| 111 |
-
return await ConvertOpenai(client, request, models, stream);
|
| 112 |
-
}
|
| 113 |
-
|
| 114 |
-
async function messagesProcess(messages) {
|
| 115 |
-
let rules = '';
|
| 116 |
-
let message = '';
|
| 117 |
-
|
| 118 |
-
for (const msg of messages) {
|
| 119 |
-
let role = msg.role;
|
| 120 |
-
// 格式化为字符串
|
| 121 |
-
const contentStr = Array.isArray(msg.content)
|
| 122 |
-
? msg.content
|
| 123 |
-
.filter((item) => item.text)
|
| 124 |
-
.map((item) => item.text)
|
| 125 |
-
.join('') || ''
|
| 126 |
-
: msg.content;
|
| 127 |
-
// 判断身份
|
| 128 |
-
if (role === 'system') {
|
| 129 |
-
rules += `system:${contentStr};\r\n`;
|
| 130 |
-
} else if (['user', 'assistant'].includes(role)) {
|
| 131 |
-
message += `${role}:${contentStr};\r\n`;
|
| 132 |
-
}
|
| 133 |
-
}
|
| 134 |
-
|
| 135 |
-
return { rules, message };
|
| 136 |
-
}
|
| 137 |
-
|
| 138 |
-
async function ConvertOpenai(client, request, model, stream) {
|
| 139 |
-
for (let i = 0; i < config.MAX_RETRY_COUNT; i++) {
|
| 140 |
-
try {
|
| 141 |
-
if (stream) {
|
| 142 |
-
const call = client.PredictWithStream(request);
|
| 143 |
-
const encoder = new TextEncoder();
|
| 144 |
-
const ReturnStream = new ReadableStream({
|
| 145 |
-
start(controller) {
|
| 146 |
-
call.on('data', (response) => {
|
| 147 |
-
let response_code = Number(response.response_code);
|
| 148 |
-
if (response_code === 204) {
|
| 149 |
-
// 如果 response_code 是 204,关闭流
|
| 150 |
-
controller.close()
|
| 151 |
-
call.destroy()
|
| 152 |
-
} else if (response_code === 200) {
|
| 153 |
-
let response_message
|
| 154 |
-
if (model.includes('gpt')) {
|
| 155 |
-
response_message = response.body.message_warpper.message.message;
|
| 156 |
-
} else {
|
| 157 |
-
response_message = response.args.args.args.message;
|
| 158 |
-
}
|
| 159 |
-
// 否则,将数据块加入流中
|
| 160 |
-
controller.enqueue(encoder.encode(`data: ${JSON.stringify(ChatCompletionStreamWithModel(response_message, model))}\n\n`));
|
| 161 |
-
} else {
|
| 162 |
-
controller.error(new Error(`Error: stream chunk is not success`));
|
| 163 |
-
controller.close()
|
| 164 |
-
}
|
| 165 |
-
})
|
| 166 |
-
}
|
| 167 |
-
});
|
| 168 |
-
return new Response(ReturnStream, {
|
| 169 |
-
headers: {
|
| 170 |
-
'Content-Type': 'text/event-stream',
|
| 171 |
-
},
|
| 172 |
-
})
|
| 173 |
-
} else {
|
| 174 |
-
const call = await new Promise((resolve, reject) => {
|
| 175 |
-
client.Predict(request, (err, response) => {
|
| 176 |
-
if (err) reject(err);
|
| 177 |
-
else resolve(response);
|
| 178 |
-
});
|
| 179 |
-
});
|
| 180 |
-
let response_code = Number(call.response_code);
|
| 181 |
-
if (response_code === 200) {
|
| 182 |
-
let response_message
|
| 183 |
-
if (model.includes('gpt')) {
|
| 184 |
-
response_message = call.body.message_warpper.message.message;
|
| 185 |
-
} else {
|
| 186 |
-
response_message = call.args.args.args.message;
|
| 187 |
-
}
|
| 188 |
-
return new Response(JSON.stringify(ChatCompletionWithModel(response_message, model)), {
|
| 189 |
-
headers: {
|
| 190 |
-
'Content-Type': 'application/json',
|
| 191 |
-
},
|
| 192 |
-
});
|
| 193 |
-
}
|
| 194 |
-
}
|
| 195 |
-
} catch (err) {
|
| 196 |
-
console.error(err);
|
| 197 |
-
await new Promise((resolve) => setTimeout(resolve, config.RETRY_DELAY));
|
| 198 |
-
}
|
| 199 |
-
}
|
| 200 |
-
return error(500, err.message);
|
| 201 |
-
}
|
| 202 |
-
|
| 203 |
-
function ChatCompletionWithModel(message, model) {
|
| 204 |
-
return {
|
| 205 |
-
id: 'Chat-Nekohy',
|
| 206 |
-
object: 'chat.completion',
|
| 207 |
-
created: Date.now(),
|
| 208 |
-
model,
|
| 209 |
-
usage: {
|
| 210 |
-
prompt_tokens: 0,
|
| 211 |
-
completion_tokens: 0,
|
| 212 |
-
total_tokens: 0,
|
| 213 |
-
},
|
| 214 |
-
choices: [
|
| 215 |
-
{
|
| 216 |
-
message: {
|
| 217 |
-
content: message,
|
| 218 |
-
role: 'assistant',
|
| 219 |
-
},
|
| 220 |
-
index: 0,
|
| 221 |
-
},
|
| 222 |
-
],
|
| 223 |
-
};
|
| 224 |
-
}
|
| 225 |
-
|
| 226 |
-
function ChatCompletionStreamWithModel(text, model) {
|
| 227 |
-
return {
|
| 228 |
-
id: 'chatcmpl-QXlha2FBbmROaXhpZUFyZUF3ZXNvbWUK',
|
| 229 |
-
object: 'chat.completion.chunk',
|
| 230 |
-
created: 0,
|
| 231 |
-
model,
|
| 232 |
-
choices: [
|
| 233 |
-
{
|
| 234 |
-
index: 0,
|
| 235 |
-
delta: {
|
| 236 |
-
content: text,
|
| 237 |
-
},
|
| 238 |
-
finish_reason: null,
|
| 239 |
-
},
|
| 240 |
-
],
|
| 241 |
-
};
|
| 242 |
-
}
|
| 243 |
-
|
| 244 |
-
async function handleCompletion(request) {
|
| 245 |
-
try {
|
| 246 |
-
// todo stream逆向接口
|
| 247 |
-
// 解析openai格式API请求
|
| 248 |
-
const { model: inputModel, messages, stream, temperature, top_p } = await request.json();
|
| 249 |
-
console.log(inputModel, messages, stream)
|
| 250 |
-
// 解析system和user/assistant消息
|
| 251 |
-
const { rules, message: content } = await messagesProcess(messages);
|
| 252 |
-
console.log(rules, content)
|
| 253 |
-
// 响应码,回复的消息
|
| 254 |
-
return await GrpcToPieces(inputModel, content, rules, stream, temperature, top_p);
|
| 255 |
-
} catch (err) {
|
| 256 |
-
return error(500, err.message);
|
| 257 |
-
}
|
| 258 |
-
}
|
| 259 |
-
|
| 260 |
-
(async () => {
|
| 261 |
-
//For Cloudflare Workers
|
| 262 |
-
if (typeof addEventListener === 'function') return;
|
| 263 |
-
// For Nodejs
|
| 264 |
-
const ittyServer = createServerAdapter(router.fetch);
|
| 265 |
-
console.log(`Listening on http://localhost:${config.PORT}`);
|
| 266 |
-
const httpServer = createServer(ittyServer);
|
| 267 |
-
httpServer.listen(config.PORT);
|
| 268 |
-
})();
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|