gitea/FastGPT
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

perf: bill

Browse Source
This commit is contained in:
archer
2023-07-13 22:53:44 +08:00
parent 726de0396b
commit f3715731c4
67 changed files with 915 additions and 1254 deletions
Download Patch File Download Diff File Expand all files Collapse all files

114
client/src/service/api/axios.ts Normal file
View File

@@ -0,0 +1,114 @@
import axios, { Method, InternalAxiosRequestConfig, AxiosResponse } from 'axios';
interface ConfigType {
headers?: { [key: string]: string };
hold?: boolean;
timeout?: number;
}
interface ResponseDataType {
code: number;
message: string;
data: any;
}
/**
* 请求开始
*/
function requestStart(config: InternalAxiosRequestConfig): InternalAxiosRequestConfig {
if (config.headers) {
// config.headers.Authorization = getToken();
}
return config;
}
/**
* 请求成功,检查请求头
*/
function responseSuccess(response: AxiosResponse<ResponseDataType>) {
return response;
}
/**
* 响应数据检查
*/
function checkRes(data: ResponseDataType) {
if (data === undefined) {
console.log('error->', data, 'data is empty');
return Promise.reject('服务器异常');
} else if (data.code < 200 || data.code >= 400) {
return Promise.reject(data);
}
return data.data;
}
/**
* 响应错误
*/
function responseError(err: any) {
console.log('error->', '请求错误', err);
if (!err) {
return Promise.reject({ message: '未知错误' });
}
if (typeof err === 'string') {
return Promise.reject({ message: err });
}
return Promise.reject(err);
}
/* 创建请求实例 */
const instance = axios.create({
timeout: 60000, // 超时时间
headers: {
'content-type': 'application/json'
}
});
/* 请求拦截 */
instance.interceptors.request.use(requestStart, (err) => Promise.reject(err));
/* 响应拦截 */
instance.interceptors.response.use(responseSuccess, (err) => Promise.reject(err));
function request(url: string, data: any, config: ConfigType, method: Method): any {
/* 去空 */
for (const key in data) {
if (data[key] === null || data[key] === undefined) {
delete data[key];
}
}
return instance
.request({
baseURL: `http://localhost:${process.env.PORT || 3000}/api`,
url,
method,
data: ['POST', 'PUT'].includes(method) ? data : null,
params: !['POST', 'PUT'].includes(method) ? data : null,
...config // 用户自定义配置，可以覆盖前面的配置
})
.then((res) => checkRes(res.data))
.catch((err) => responseError(err));
}
/**
* api请求方式
* @param {String} url
* @param {Any} params
* @param {Object} config
* @returns
*/
export function GET<T>(url: string, params = {}, config: ConfigType = {}): Promise<T> {
return request(url, params, config, 'GET');
}
export function POST<T>(url: string, data = {}, config: ConfigType = {}): Promise<T> {
return request(url, data, config, 'POST');
}
export function PUT<T>(url: string, data = {}, config: ConfigType = {}): Promise<T> {
return request(url, data, config, 'PUT');
}
export function DELETE<T>(url: string, data = {}, config: ConfigType = {}): Promise<T> {
return request(url, data, config, 'DELETE');
}

2
client/src/service/api/request.ts
View File

@@ -93,6 +93,8 @@ export const moduleFetch = ({ url, data, res }: Props) =>
event: sseResponseEventEnum.answer,
data: JSON.stringify(data)
});
} else if (item.event === sseResponseEventEnum.error) {
return reject(getErrText(data, '流响应错误'));
}
});
read();

82
client/src/service/events/generateQA.ts
View File

@@ -1,15 +1,16 @@
import { TrainingData } from '@/service/mongo';
import { getApiKey } from '../utils/auth';
import { OpenAiChatEnum } from '@/constants/model';
import { pushSplitDataBill } from '@/service/events/pushBill';
import { openaiAccountError } from '../errorCode';
import { modelServiceToolMap } from '../utils/chat';
import { ChatRoleEnum } from '@/constants/chat';
import { BillTypeEnum } from '@/constants/user';
import { BillSourceEnum } from '@/constants/user';
import { pushDataToKb } from '@/pages/api/openapi/kb/pushData';
import { TrainingModeEnum } from '@/constants/plugin';
import { ERROR_ENUM } from '../errorCode';
import { sendInform } from '@/pages/api/user/inform/send';
import { authBalanceByUid } from '../utils/auth';
import { axiosConfig, getOpenAIApi } from '../ai/openai';
import { ChatCompletionRequestMessage } from 'openai';
const reduceQueue = () => {
global.qaQueueLen = global.qaQueueLen > 0 ? global.qaQueueLen - 1 : 0;
@@ -37,7 +38,8 @@ export async function generateQA(): Promise<any> {
kbId: 1,
prompt: 1,
q: 1,
source: 1
source: 1,
model: 1
});
// task preemption
@@ -51,54 +53,59 @@ export async function generateQA(): Promise<any> {
userId = String(data.userId);
const kbId = String(data.kbId);
// 余额校验并获取 openapi Key
const { systemAuthKey } = await getApiKey({
model: OpenAiChatEnum.GPT35,
userId,
mustPay: true
});
await authBalanceByUid(userId);
const startTime = Date.now();
const chatAPI = getOpenAIApi();
// 请求 chatgpt 获取回答
const response = await Promise.all(
[data.q].map((text) =>
modelServiceToolMap
.chatCompletion({
model: OpenAiChatEnum.GPT3516k,
apiKey: systemAuthKey,
temperature: 0.8,
messages: [
{
obj: ChatRoleEnum.System,
value: `你是出题人.
[data.q].map((text) => {
const messages: ChatCompletionRequestMessage[] = [
{
role: 'system',
content: `你是出题人.
${data.prompt || '用户会发送一段长文本'}.
从中选出 25 个问题和答案. 答案详细完整. 按格式回答: Q1:
A1:
Q2:
A2:
...`
},
{
obj: 'Human',
value: text
}
],
stream: false
})
.then(({ totalTokens, responseText, responseMessages }) => {
const result = formatSplitText(responseText); // 格式化后的QA对
},
{
role: 'user',
content: text
}
];
return chatAPI
.createChatCompletion(
{
model: data.model,
temperature: 0.8,
messages,
stream: false
},
{
timeout: 480000,
...axiosConfig()
}
)
.then((res) => {
const answer = res.data.choices?.[0].message?.content;
const totalTokens = res.data.usage?.total_tokens || 0;
const result = formatSplitText(answer || ''); // 格式化后的QA对
console.log(`split result length: `, result.length);
// 计费
pushSplitDataBill({
isPay: result.length > 0,
userId: data.userId,
type: BillTypeEnum.QA,
textLen: responseMessages.map((item) => item.value).join('').length,
totalTokens
totalTokens,
model: data.model,
appName: 'QA 拆分'
});
return {
rawContent: responseText,
rawContent: answer,
result
};
})
@@ -106,8 +113,8 @@ A2:
console.log('QA拆分错误');
console.log(err.response?.status, err.response?.statusText, err.response?.data);
return Promise.reject(err);
})
)
});
})
);
const responseList = response.map((item) => item.result).flat();
@@ -120,6 +127,7 @@ A2:
source: data.source
})),
userId,
model: global.vectorModels[0].model,
mode: TrainingModeEnum.index
});

11
client/src/service/events/generateVector.ts
View File

@@ -1,6 +1,6 @@
import { openaiAccountError } from '../errorCode';
import { insertKbItem } from '@/service/pg';
import { openaiEmbedding } from '@/pages/api/openapi/plugin/openaiEmbedding';
import { getVector } from '@/pages/api/openapi/plugin/vector';
import { TrainingData } from '../models/trainingData';
import { ERROR_ENUM } from '../errorCode';
import { TrainingModeEnum } from '@/constants/plugin';
@@ -33,7 +33,8 @@ export async function generateVector(): Promise<any> {
kbId: 1,
q: 1,
a: 1,
source: 1
source: 1,
model: 1
});
// task preemption
@@ -55,10 +56,10 @@ export async function generateVector(): Promise<any> {
];
// 生成词向量
const vectors = await openaiEmbedding({
const vectors = await getVector({
model: data.model,
input: dataItems.map((item) => item.q),
userId,
mustPay: true
userId
});
// 生成结果插入到 pg

188
client/src/service/events/pushBill.ts
View File

@@ -1,66 +1,85 @@
import { connectToDatabase, Bill, User, ShareChat } from '../mongo';
import {
ChatModelMap,
OpenAiChatEnum,
ChatModelType,
embeddingModel,
embeddingPrice
} from '@/constants/model';
import { BillTypeEnum } from '@/constants/user';
import { BillSourceEnum } from '@/constants/user';
import { getModel } from '../utils/data';
import type { BillListItemType } from '@/types/mongoSchema';
export const pushChatBill = async ({
isPay,
chatModel,
userId,
export const createTaskBill = async ({
appName,
appId,
textLen,
tokens,
type
userId,
source
}: {
isPay: boolean;
chatModel: ChatModelType;
userId: string;
appName: string;
appId: string;
textLen: number;
tokens: number;
type: BillTypeEnum.chat | BillTypeEnum.openapiChat;
userId: string;
source: `${BillSourceEnum}`;
}) => {
console.log(`chat generate success. text len: ${textLen}. token len: ${tokens}. pay:${isPay}`);
if (!isPay) return;
const res = await Bill.create({
userId,
appName,
appId,
total: 0,
source,
list: []
});
return String(res._id);
};
let billId = '';
export const pushTaskBillListItem = async ({
billId,
moduleName,
amount,
model,
tokenLen
}: { billId?: string } & BillListItemType) => {
if (!billId) return;
try {
await Bill.findByIdAndUpdate(billId, {
$push: {
list: {
moduleName,
amount,
model,
tokenLen
}
}
});
} catch (error) {}
};
export const finishTaskBill = async ({ billId }: { billId: string }) => {
try {
// update bill
const res = await Bill.findByIdAndUpdate(billId, [
{
$set: {
total: {
$sum: '$list.amount'
},
time: new Date()
}
}
]);
if (!res) return;
const total = res.list.reduce((sum, item) => sum + item.amount, 0) || 0;
console.log('finish bill:', total);
// 账号扣费
await User.findByIdAndUpdate(res.userId, {
$inc: { balance: -total }
});
} catch (error) {
console.log('Finish bill failed:', error);
billId && Bill.findByIdAndDelete(billId);
}
};
export const delTaskBill = async (billId?: string) => {
if (!billId) return;
try {
await connectToDatabase();
// 计算价格
const unitPrice = ChatModelMap[chatModel]?.price || 3;
const price = unitPrice * tokens;
try {
// 插入 Bill 记录
const res = await Bill.create({
userId,
type,
modelName: chatModel,
appId,
textLen,
tokenLen: tokens,
price
});
billId = res._id;
// 账号扣费
await User.findByIdAndUpdate(userId, {
$inc: { balance: -price }
});
} catch (error) {
console.log('创建账单失败:', error);
billId && Bill.findByIdAndDelete(billId);
}
} catch (error) {
console.log(error);
}
await Bill.findByIdAndRemove(billId);
} catch (error) {}
};
export const updateShareChatBill = async ({
@@ -81,22 +100,17 @@ export const updateShareChatBill = async ({
};
export const pushSplitDataBill = async ({
isPay,
userId,
totalTokens,
textLen,
type
model,
appName
}: {
isPay: boolean;
model: string;
userId: string;
totalTokens: number;
textLen: number;
type: BillTypeEnum.QA;
appName: string;
}) => {
console.log(
`splitData generate success. text len: ${textLen}. token len: ${totalTokens}. pay:${isPay}`
);
if (!isPay) return;
console.log(`splitData generate success. token len: ${totalTokens}.`);
let billId;
@@ -104,24 +118,22 @@ export const pushSplitDataBill = async ({
await connectToDatabase();
// 获取模型单价格, 都是用 gpt35 拆分
const unitPrice = ChatModelMap[OpenAiChatEnum.GPT3516k].price || 3;
const unitPrice = global.chatModels.find((item) => item.model === model)?.price || 3;
// 计算价格
const price = unitPrice * totalTokens;
const total = unitPrice * totalTokens;
// 插入 Bill 记录
const res = await Bill.create({
userId,
type,
modelName: OpenAiChatEnum.GPT3516k,
textLen,
appName,
tokenLen: totalTokens,
price
total
});
billId = res._id;
// 账号扣费
await User.findByIdAndUpdate(userId, {
$inc: { balance: -price }
$inc: { balance: -total }
});
} catch (error) {
console.log('创建账单失败:', error);
@@ -130,21 +142,14 @@ export const pushSplitDataBill = async ({
};
export const pushGenerateVectorBill = async ({
isPay,
userId,
text,
tokenLen
tokenLen,
model
}: {
isPay: boolean;
userId: string;
text: string;
tokenLen: number;
model: string;
}) => {
// console.log(
// `vector generate success. text len: ${text.length}. token len: ${tokenLen}. pay:${isPay}`
// );
if (!isPay) return;
let billId;
try {
@@ -152,23 +157,22 @@ export const pushGenerateVectorBill = async ({
try {
// 计算价格. 至少为1
let price = embeddingPrice * tokenLen;
price = price > 1 ? price : 1;
const unitPrice = global.vectorModels.find((item) => item.model === model)?.price || 0.2;
let total = unitPrice * tokenLen;
total = total > 1 ? total : 1;
// 插入 Bill 记录
const res = await Bill.create({
userId,
type: BillTypeEnum.vector,
modelName: embeddingModel,
textLen: text.length,
tokenLen,
price
model,
appName: '索引生成',
total
});
billId = res._id;
// 账号扣费
await User.findByIdAndUpdate(userId, {
$inc: { balance: -price }
$inc: { balance: -total }
});
} catch (error) {
console.log('创建账单失败:', error);
@@ -178,3 +182,9 @@ export const pushGenerateVectorBill = async ({
console.log(error);
}
};
export const countModelPrice = ({ model, tokens }: { model: string; tokens: number }) => {
const modelData = getModel(model);
if (!modelData) return 0;
return modelData.price * tokens;
};

45
client/src/service/models/app.ts
View File

@@ -1,6 +1,5 @@
import { Schema, model, models, Model } from 'mongoose';
import { AppSchema as AppType } from '@/types/mongoSchema';
import { ChatModelMap, OpenAiChatEnum } from '@/constants/model';
const AppSchema = new Schema({
userId: {
@@ -24,50 +23,6 @@ const AppSchema = new Schema({
type: Date,
default: () => new Date()
},
chat: {
relatedKbs: {
type: [Schema.Types.ObjectId],
ref: 'kb',
default: []
},
searchSimilarity: {
type: Number,
default: 0.8
},
searchLimit: {
type: Number,
default: 5
},
searchEmptyText: {
type: String,
default: ''
},
systemPrompt: {
type: String,
default: ''
},
limitPrompt: {
type: String,
default: ''
},
maxToken: {
type: Number,
default: 4000,
min: 100
},
temperature: {
type: Number,
min: 0,
max: 10,
default: 0
},
chatModel: {
// 聊天时使用的模型
type: String,
enum: Object.keys(ChatModelMap),
default: OpenAiChatEnum.GPT3516k
}
},
share: {
topNum: {
type: Number,

49
client/src/service/models/bill.ts
View File

@@ -1,7 +1,6 @@
import { Schema, model, models, Model } from 'mongoose';
import { ChatModelMap, embeddingModel } from '@/constants/model';
import { BillSchema as BillType } from '@/types/mongoSchema';
import { BillTypeMap } from '@/constants/user';
import { BillSourceEnum, BillSourceMap } from '@/constants/user';
const BillSchema = new Schema({
userId: {
@@ -9,36 +8,48 @@ const BillSchema = new Schema({
ref: 'user',
required: true
},
type: {
appName: {
type: String,
enum: Object.keys(BillTypeMap),
required: true
},
modelName: {
type: String,
enum: [...Object.keys(ChatModelMap), embeddingModel]
default: ''
},
appId: {
type: Schema.Types.ObjectId,
ref: 'app'
ref: 'app',
required: false
},
time: {
type: Date,
default: () => new Date()
},
textLen: {
// 提示词+响应的总字数
total: {
type: Number,
required: true
},
tokenLen: {
// 折算成 token 的数量
type: Number,
required: true
source: {
type: String,
enum: Object.keys(BillSourceMap),
default: BillSourceEnum.fastgpt
},
price: {
type: Number,
required: true
list: {
type: [
{
moduleName: {
type: String,
required: true
},
amount: {
type: Number,
required: true
},
model: {
type: String
},
tokenLen: {
type: Number
}
}
],
default: []
}
});

22
client/src/service/models/system.ts
View File

@@ -1,22 +0,0 @@
import { Schema, model, models } from 'mongoose';
const SystemSchema = new Schema({
vectorMaxProcess: {
type: Number,
default: 10
},
qaMaxProcess: {
type: Number,
default: 10
},
pgIvfflatProbe: {
type: Number,
default: 10
},
sensitiveCheck: {
type: Boolean,
default: false
}
});
export const System = models['system'] || model('system', SystemSchema);

7
client/src/service/models/trainingData.ts
View File

@@ -28,13 +28,16 @@ const TrainingDataSchema = new Schema({
enum: Object.keys(TrainingTypeMap),
required: true
},
model: {
type: String,
required: true
},
prompt: {
// 拆分时的提示词
// qa split prompt
type: String,
default: ''
},
q: {
// 如果是
type: String,
default: ''
},

7
client/src/service/mongo.ts
View File

@@ -2,6 +2,7 @@ import mongoose from 'mongoose';
import tunnel from 'tunnel';
import { startQueue } from './utils/tools';
import { updateSystemEnv } from '@/pages/api/system/updateEnv';
import { initSystemModels } from '@/pages/api/system/getInitData';
/**
* 连接 MongoDB 数据库
@@ -10,6 +11,7 @@ export async function connectToDatabase(): Promise<void> {
if (global.mongodb) {
return;
}
global.mongodb = 'connecting';
// init global data
global.qaQueueLen = 0;
@@ -31,8 +33,9 @@ export async function connectToDatabase(): Promise<void> {
}
});
}
initSystemModels();
updateSystemEnv();
global.mongodb = 'connecting';
try {
mongoose.set('strictQuery', true);
global.mongodb = await mongoose.connect(process.env.MONGODB_URI as string, {
@@ -49,7 +52,6 @@ export async function connectToDatabase(): Promise<void> {
}
// init function
updateSystemEnv();
startQueue();
}
@@ -66,5 +68,4 @@ export * from './models/collection';
export * from './models/shareChat';
export * from './models/kb';
export * from './models/inform';
export * from './models/system';
export * from './models/image';

2
client/src/service/response.ts
View File

@@ -92,7 +92,7 @@ export const sseErrRes = (res: NextApiResponse, error: any) => {
} else if (openaiError[error?.response?.statusText]) {
msg = openaiError[error.response.statusText];
}
console.log('sse error', error);
console.log('sse error => ', error);
sseResponse({
res,

73
client/src/service/utils/auth.ts
View File

@@ -1,15 +1,11 @@
import type { NextApiRequest } from 'next';
import jwt from 'jsonwebtoken';
import Cookie from 'cookie';
import { Chat, App, OpenApi, User, ShareChat, KB } from '../mongo';
import { App, OpenApi, User, ShareChat, KB } from '../mongo';
import type { AppSchema } from '@/types/mongoSchema';
import type { ChatItemType } from '@/types/chat';
import mongoose from 'mongoose';
import { defaultApp } from '@/constants/model';
import { formatPrice } from '@/utils/user';
import { ERROR_ENUM } from '../errorCode';
import { ChatModelType, OpenAiChatEnum } from '@/constants/model';
import { hashPassword } from '@/service/utils/tools';
export type AuthType = 'token' | 'root' | 'apikey';
@@ -35,6 +31,19 @@ export const parseCookie = (cookie?: string): Promise<string> => {
});
};
/* auth balance */
export const authBalanceByUid = async (uid: string) => {
const user = await User.findById(uid);
if (!user) {
return Promise.reject(ERROR_ENUM.unAuthorization);
}
if (!user.openaiKey && formatPrice(user.balance) <= 0) {
return Promise.reject(ERROR_ENUM.insufficientQuota);
}
return user;
};
/* uniform auth user */
export const authUser = async ({
req,
@@ -144,14 +153,7 @@ export const authUser = async ({
// balance check
if (authBalance) {
const user = await User.findById(uid);
if (!user) {
return Promise.reject(ERROR_ENUM.unAuthorization);
}
if (!user.openaiKey && formatPrice(user.balance) <= 0) {
return Promise.reject(ERROR_ENUM.insufficientQuota);
}
await authBalanceByUid(uid);
}
return {
@@ -166,43 +168,6 @@ export const getSystemOpenAiKey = () => {
return process.env.ONEAPI_KEY || process.env.OPENAIKEY || '';
};
/* 获取 api 请求的 key */
export const getApiKey = async ({
model,
userId,
mustPay = false
}: {
model: ChatModelType;
userId: string;
mustPay?: boolean;
}) => {
const user = await User.findById(userId, 'openaiKey balance');
if (!user) {
return Promise.reject(ERROR_ENUM.unAuthorization);
}
const userOpenAiKey = user.openaiKey || '';
const systemAuthKey = getSystemOpenAiKey();
// 有自己的key
if (!mustPay && userOpenAiKey) {
return {
userOpenAiKey,
systemAuthKey: ''
};
}
// 平台账号余额校验
if (formatPrice(user.balance) <= 0) {
return Promise.reject(ERROR_ENUM.insufficientQuota);
}
return {
userOpenAiKey: '',
systemAuthKey
};
};
// 模型使用权校验
export const authApp = async ({
appId,
@@ -232,14 +197,6 @@ export const authApp = async ({
if (userId !== String(app.userId)) return Promise.reject(ERROR_ENUM.unAuthModel);
}
// do not share detail info
if (!reserveDetail && !app.share.isShareDetail && userId !== String(app.userId)) {
app.chat = {
...defaultApp.chat,
chatModel: app.chat.chatModel
};
}
return {
app,
showModelDetail: userId === String(app.userId)

137
client/src/service/utils/chat/index.ts
View File

@@ -1,13 +1,8 @@
import { ChatItemType } from '@/types/chat';
import { modelToolMap } from '@/utils/plugin';
import type { ChatModelType } from '@/constants/model';
import { ChatRoleEnum, sseResponseEventEnum } from '@/constants/chat';
import { sseResponse } from '../tools';
import { ChatRoleEnum } from '@/constants/chat';
import { OpenAiChatEnum } from '@/constants/model';
import { chatResponse, openAiStreamResponse } from './openai';
import type { NextApiResponse } from 'next';
import { textAdaptGptResponse } from '@/utils/adapt';
import { parseStreamChunk } from '@/utils/adapt';
export type ChatCompletionType = {
apiKey: string;
@@ -36,11 +31,6 @@ export type StreamResponseReturnType = {
finishMessages: ChatItemType[];
};
export const modelServiceToolMap = {
chatCompletion: chatResponse,
streamResponse: openAiStreamResponse
};
/* delete invalid symbol */
const simplifyStr = (str = '') =>
str
@@ -54,7 +44,7 @@ export const ChatContextFilter = ({
prompts,
maxTokens
}: {
model: ChatModelType;
model: string;
prompts: ChatItemType[];
maxTokens: number;
}) => {
@@ -111,126 +101,3 @@ export const ChatContextFilter = ({
return [...systemPrompts, ...chats];
};
/* stream response */
export const resStreamResponse = async ({
model,
res,
chatResponse,
prompts
}: StreamResponseType & {
model: ChatModelType;
}) => {
// 创建响应流
res.setHeader('Content-Type', 'text/event-stream;charset=utf-8');
res.setHeader('Access-Control-Allow-Origin', '*');
res.setHeader('X-Accel-Buffering', 'no');
res.setHeader('Cache-Control', 'no-cache, no-transform');
const { responseContent, totalTokens, finishMessages } = await modelServiceToolMap.streamResponse(
{
chatResponse,
prompts,
res,
model
}
);
return { responseContent, totalTokens, finishMessages };
};
/* stream response */
export const V2_StreamResponse = async ({
model,
res,
chatResponse,
prompts
}: StreamResponseType & {
model: ChatModelType;
}) => {
let responseContent = '';
let error: any = null;
let truncateData = '';
const clientRes = async (data: string) => {
//部分代理会导致流式传输时的数据被截断，不为json格式，这里做一个兼容
const { content = '' } = (() => {
try {
if (truncateData) {
try {
//判断是否为json，如果是的话直接跳过后续拼装操作，注意极端情况下可能出现截断成3截以上情况也可以兼容
JSON.parse(data);
} catch (e) {
data = truncateData + data;
}
truncateData = '';
}
const json = JSON.parse(data);
const content: string = json?.choices?.[0].delta.content || '';
error = json.error;
responseContent += content;
return { content };
} catch (error) {
truncateData = data;
return {};
}
})();
if (res.closed || error) return;
if (data === '[DONE]') {
sseResponse({
res,
event: sseResponseEventEnum.answer,
data: textAdaptGptResponse({
text: null,
finish_reason: 'stop'
})
});
sseResponse({
res,
event: sseResponseEventEnum.answer,
data: '[DONE]'
});
} else {
sseResponse({
res,
event: sseResponseEventEnum.answer,
data: textAdaptGptResponse({
text: content
})
});
}
};
try {
for await (const chunk of chatResponse.data as any) {
if (res.closed) break;
const parse = parseStreamChunk(chunk);
parse.forEach((item) => clientRes(item.data));
}
} catch (error) {
console.log('pipe error', error);
}
if (error) {
console.log(error);
return Promise.reject(error);
}
// count tokens
const finishMessages = prompts.concat({
obj: ChatRoleEnum.AI,
value: responseContent
});
const totalTokens = modelToolMap.countTokens({
model,
messages: finishMessages
});
return {
responseContent,
totalTokens,
finishMessages
};
};

133
client/src/service/utils/chat/openai.ts
View File

@@ -1,133 +0,0 @@
import { Configuration, OpenAIApi } from 'openai';
import { axiosConfig } from '../tools';
import { ChatModelMap, OpenAiChatEnum } from '@/constants/model';
import { adaptChatItem_openAI } from '@/utils/plugin/openai';
import { modelToolMap } from '@/utils/plugin';
import { ChatCompletionType, ChatContextFilter, StreamResponseType } from './index';
import { ChatRoleEnum } from '@/constants/chat';
import { parseStreamChunk } from '@/utils/adapt';
export const getOpenAIApi = (apiKey: string) => {
const openaiBaseUrl = process.env.OPENAI_BASE_URL || 'https://api.openai.com/v1';
return new OpenAIApi(
new Configuration({
basePath: apiKey === process.env.ONEAPI_KEY ? process.env.ONEAPI_URL : openaiBaseUrl
})
);
};
/* 模型对话 */
export const chatResponse = async ({
model,
apiKey,
temperature,
maxToken = 4000,
messages,
stream
}: ChatCompletionType & { model: `${OpenAiChatEnum}` }) => {
const modelTokenLimit = ChatModelMap[model]?.contextMaxToken || 4000;
const filterMessages = ChatContextFilter({
model,
prompts: messages,
maxTokens: Math.ceil(modelTokenLimit - 300) // filter token. not response maxToken
});
const adaptMessages = adaptChatItem_openAI({ messages: filterMessages, reserveId: false });
const chatAPI = getOpenAIApi(apiKey);
const promptsToken = modelToolMap.countTokens({
model,
messages: filterMessages
});
maxToken = maxToken + promptsToken > modelTokenLimit ? modelTokenLimit - promptsToken : maxToken;
const response = await chatAPI.createChatCompletion(
{
model,
temperature: Number(temperature || 0),
max_tokens: maxToken,
messages: adaptMessages,
frequency_penalty: 0.5, // 越大，重复内容越少
presence_penalty: -0.5, // 越大，越容易出现新内容
stream
// stop: ['.!?。']
},
{
timeout: stream ? 60000 : 480000,
responseType: stream ? 'stream' : 'json',
...axiosConfig(apiKey)
}
);
const responseText = stream ? '' : response.data.choices?.[0].message?.content || '';
const totalTokens = stream ? 0 : response.data.usage?.total_tokens || 0;
return {
streamResponse: response,
responseMessages: filterMessages.concat({ obj: 'AI', value: responseText }),
responseText,
totalTokens
};
};
/* openai stream response */
export const openAiStreamResponse = async ({
res,
model,
chatResponse,
prompts
}: StreamResponseType & {
model: `${OpenAiChatEnum}`;
}) => {
try {
let responseContent = '';
const clientRes = async (data: string) => {
const { content = '' } = (() => {
try {
const json = JSON.parse(data);
const content: string = json?.choices?.[0].delta.content || '';
responseContent += content;
return { content };
} catch (error) {
return {};
}
})();
if (data === '[DONE]') return;
!res.closed && content && res.write(content);
};
try {
for await (const chunk of chatResponse.data as any) {
if (res.closed) break;
const parse = parseStreamChunk(chunk);
parse.forEach((item) => clientRes(item.data));
}
} catch (error) {
console.log('pipe error', error);
}
// count tokens
const finishMessages = prompts.concat({
obj: ChatRoleEnum.AI,
value: responseContent
});
const totalTokens = modelToolMap.countTokens({
model,
messages: finishMessages
});
return {
responseContent,
totalTokens,
finishMessages
};
} catch (error) {
return Promise.reject(error);
}
};

14
client/src/service/utils/data.ts Normal file
View File

@@ -0,0 +1,14 @@
export const getChatModel = (model: string) => {
return global.chatModels.find((item) => item.model === model);
};
export const getVectorModel = (model: string) => {
return global.vectorModels.find((item) => item.model === model);
};
export const getQAModel = (model: string) => {
return global.qaModels.find((item) => item.model === model);
};
export const getModel = (model: string) => {
return [...global.chatModels, ...global.vectorModels, ...global.qaModels].find(
(item) => item.model === model
);
};

15
client/src/service/utils/tools.ts
View File

@@ -4,7 +4,6 @@ import crypto from 'crypto';
import jwt from 'jsonwebtoken';
import { generateQA } from '../events/generateQA';
import { generateVector } from '../events/generateVector';
import { sseResponseEventEnum } from '@/constants/chat';
/* 密码加密 */
export const hashPassword = (psw: string) => {
@@ -33,20 +32,6 @@ export const clearCookie = (res: NextApiResponse) => {
res.setHeader('Set-Cookie', 'token=; Path=/; Max-Age=0');
};
/* openai axios config */
export const axiosConfig = (apikey: string) => {
const openaiBaseUrl = process.env.OPENAI_BASE_URL || 'https://api.openai.com/v1';
return {
baseURL: apikey === process.env.ONEAPI_KEY ? process.env.ONEAPI_URL : openaiBaseUrl, // 此处仅对非 npm 模块有效
httpsAgent: global.httpsAgent,
headers: {
Authorization: `Bearer ${apikey}`,
auth: process.env.OPENAI_BASE_URL_AUTH || ''
}
};
};
export function withNextCors(handler: NextApiHandler): NextApiHandler {
return async function nextApiHandlerWrappedWithNextCors(
req: NextApiRequest,