perf: code and inform

client/public/docs/chatProblem.md

@@ -7,7 +7,7 @@
 | 计费项 | 价格: 元/ 1K tokens（包含上下文）|
 | --- | --- |
 | 知识库 - 索引 | 0.001 |
-| chatgpt - 对话 | 0.025 |
+| chatgpt - 对话 | 0.015 |
 | chatgpt16K - 对话 | 0.025 |
 | gpt4 - 对话 | 0.45 |
 | 文件拆分 | 0.025 |

client/public/docs/intro.md

@@ -19,7 +19,7 @@ FastGpt 项目完全开源，可随意私有化部署，去除平台风险忧虑
 | 计费项 | 价格: 元/ 1K tokens（包含上下文）|
 | --- | --- |
 | 知识库 - 索引 | 0.001 |
-| chatgpt - 对话 | 0.025 |
+| chatgpt - 对话 | 0.015 |
 | chatgpt16K - 对话 | 0.025 |
 | gpt4 - 对话 | 0.45 |
 | 文件拆分 | 0.025 |

client/src/constants/model.ts

@@ -31,7 +31,7 @@ export const ChatModelMap = {
     contextMaxToken: 4000,
     systemMaxToken: 2400,
     maxTemperature: 1.2,
-    price: 2.5
+    price: 1.5
   },
   [OpenAiChatEnum.GPT3516k]: {
     chatModel: OpenAiChatEnum.GPT3516k,

client/src/pages/api/openapi/chat/chat.ts

@@ -129,7 +129,8 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
 
     // 发出请求
     const { streamResponse, responseMessages, responseText, totalTokens } =
-      await modelServiceToolMap[model.chat.chatModel].chatCompletion({
+      await modelServiceToolMap.chatCompletion({
+        model: model.chat.chatModel,
         apiKey,
         temperature: +temperature,
         messages: completePrompts,

client/src/pages/api/openapi/kb/appKbSearch.ts

@@ -150,13 +150,15 @@ export async function appKbSearch({
     }
   ];
 
-  const fixedSystemTokens = modelToolMap[model.chat.chatModel].countTokens({
+  const fixedSystemTokens = modelToolMap.countTokens({
+    model: model.chat.chatModel,
     messages: [...userSystemPrompt, ...userLimitPrompt]
   });
 
   // filter part quote by maxToken
-  const sliceResult = modelToolMap[model.chat.chatModel]
+  const sliceResult = modelToolMap
     .tokenSlice({
+      model: model.chat.chatModel,
       maxToken: modelConstantsData.systemMaxToken - fixedSystemTokens,
       messages: filterSearch.map((item, i) => ({
         obj: ChatRoleEnum.System,

client/src/pages/api/openapi/kb/pushData.ts

@@ -78,7 +78,8 @@ export async function pushDataToKb({
 
     if (mode === TrainingModeEnum.qa) {
       // count token
-      const token = modelToolMap[OpenAiChatEnum.GPT3516k].countTokens({
+      const token = modelToolMap.countTokens({
+        model: OpenAiChatEnum.GPT3516k,
         messages: [{ obj: 'System', value: item.q }]
       });
       if (token > modeMaxToken[TrainingModeEnum.qa]) {

client/src/pages/api/openapi/v1/chat/completions.ts

@@ -196,7 +196,8 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
 
     // start model api. responseText and totalTokens: valid only if stream = false
     const { streamResponse, responseMessages, responseText, totalTokens } =
-      await modelServiceToolMap[model.chat.chatModel].chatCompletion({
+      await modelServiceToolMap.chatCompletion({
+        model: model.chat.chatModel,
         apiKey: userOpenAiKey || apiKey,
         temperature: +temperature,
         maxToken: model.chat.maxToken,

client/src/pages/api/system/getModels.ts

@@ -9,6 +9,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
 
   chatModelList.push(ChatModelMap[OpenAiChatEnum.GPT3516k]);
   chatModelList.push(ChatModelMap[OpenAiChatEnum.GPT35]);
+  chatModelList.push(ChatModelMap[OpenAiChatEnum.textada001]);
   chatModelList.push(ChatModelMap[OpenAiChatEnum.GPT4]);
 
   jsonRes(res, {

client/src/pages/api/user/inform/send.ts

@@ -4,6 +4,7 @@ import { jsonRes } from '@/service/response';
 import { connectToDatabase, Inform, User } from '@/service/mongo';
 import { authUser } from '@/service/utils/auth';
 import { InformTypeEnum } from '@/constants/user';
+import { startSendInform } from '@/service/events/sendInform';
 
 export type Props = {
   type: `${InformTypeEnum}`;
@@ -37,25 +38,26 @@ export async function sendInform({ type, title, content, userId }: Props) {
 
   try {
     if (userId) {
-      // skip it if have same inform within 5 minutes
+      global.sendInformQueue.push(async () => {
-      const inform = await Inform.findOne({
+        // skip it if have same inform within 5 minutes
-        type,
+        const inform = await Inform.findOne({
-        title,
+          type,
-        content,
+          title,
-        userId,
+          content,
-        read: false,
+          userId,
-        time: { $lte: new Date(Date.now() + 5 * 60 * 1000) }
+          time: { $gte: new Date(Date.now() - 5 * 60 * 1000) }
+        });
+
+        if (inform) return;
+
+        await Inform.create({
+          type,
+          title,
+          content,
+          userId
+        });
       });
-
+      startSendInform();
-      if (inform) return;
-
-      await Inform.create({
-        type,
-        title,
-        content,
-        userId
-      });
-
       return;
     }
 

client/src/pages/number/components/PayModal.tsx

@@ -114,7 +114,7 @@ const PayModal = ({ onClose }: { onClose: () => void }) => {
 | 计费项 | 价格: 元/ 1K tokens(包含上下文)|
 | --- | --- |
 | 知识库 - 索引 | 0.001 |
-| chatgpt - 对话 | 0.025 |
+| chatgpt - 对话 | 0.015 |
 | chatgpt16K - 对话 | 0.025 |
 | gpt4 - 对话 | 0.45 |
 | 文件拆分 | 0.025 |`}

client/src/service/events/generateQA.ts

@@ -63,8 +63,9 @@ export async function generateQA(): Promise<any> {
     // 请求 chatgpt 获取回答
     const response = await Promise.all(
       [data.q].map((text) =>
-        modelServiceToolMap[OpenAiChatEnum.GPT3516k]
+        modelServiceToolMap
           .chatCompletion({
+            model: OpenAiChatEnum.GPT3516k,
             apiKey: systemAuthKey,
             temperature: 0.8,
             messages: [

client/src/service/events/sendInform.ts

@@ -0,0 +1,16 @@
+export const startSendInform = async () => {
+  if (global.sendInformQueue.length === 0 || global.sendInformQueueLen > 0) return;
+  global.sendInformQueueLen++;
+
+  try {
+    const fn = global.sendInformQueue[global.sendInformQueue.length - 1];
+    await fn();
+    global.sendInformQueue.pop();
+    global.sendInformQueueLen--;
+
+    startSendInform();
+  } catch (error) {
+    global.sendInformQueueLen--;
+    startSendInform();
+  }
+};

client/src/service/mongo.ts

@@ -20,6 +20,8 @@ export async function connectToDatabase(): Promise<void> {
     pgIvfflatProbe: 10,
     sensitiveCheck: false
   };
+  global.sendInformQueue = [];
+  global.sendInformQueueLen = 0;
   // proxy obj
   if (process.env.AXIOS_PROXY_HOST && process.env.AXIOS_PROXY_PORT) {
     global.httpsAgent = tunnel.httpsOverHttp({

client/src/service/utils/auth.ts

@@ -181,33 +181,13 @@ export const getApiKey = async ({
     return Promise.reject(ERROR_ENUM.unAuthorization);
   }
 
-  const keyMap = {
+  const userOpenAiKey = user.openaiKey || '';
-    [OpenAiChatEnum.GPT35]: {
+  const systemAuthKey = getSystemOpenAiKey();
-      userOpenAiKey: user.openaiKey || '',
-      systemAuthKey: getSystemOpenAiKey()
-    },
-    [OpenAiChatEnum.GPT3516k]: {
-      userOpenAiKey: user.openaiKey || '',
-      systemAuthKey: getSystemOpenAiKey()
-    },
-    [OpenAiChatEnum.GPT4]: {
-      userOpenAiKey: user.openaiKey || '',
-      systemAuthKey: getSystemOpenAiKey()
-    },
-    [OpenAiChatEnum.GPT432k]: {
-      userOpenAiKey: user.openaiKey || '',
-      systemAuthKey: getSystemOpenAiKey()
-    }
-  };
-
-  if (!keyMap[model]) {
-    return Promise.reject('App  model is exists');
-  }
 
   // 有自己的key
-  if (!mustPay && keyMap[model].userOpenAiKey) {
+  if (!mustPay && userOpenAiKey) {
     return {
-      userOpenAiKey: keyMap[model].userOpenAiKey,
+      userOpenAiKey,
       systemAuthKey: ''
     };
   }
@@ -219,7 +199,7 @@ export const getApiKey = async ({
 
   return {
     userOpenAiKey: '',
-    systemAuthKey: keyMap[model].systemAuthKey
+    systemAuthKey
   };
 };
 

client/src/service/utils/chat/index.ts

@@ -27,6 +27,7 @@ export type StreamResponseType = {
   chatResponse: any;
   prompts: ChatItemType[];
   res: NextApiResponse;
+  model: `${OpenAiChatEnum}`;
   [key: string]: any;
 };
 export type StreamResponseReturnType = {
@@ -35,49 +36,9 @@ export type StreamResponseReturnType = {
   finishMessages: ChatItemType[];
 };
 
-export const modelServiceToolMap: Record<
+export const modelServiceToolMap = {
-  ChatModelType,
+  chatCompletion: chatResponse,
-  {
+  streamResponse: openAiStreamResponse
-    chatCompletion: (data: ChatCompletionType) => Promise<ChatCompletionResponseType>;
-    streamResponse: (data: StreamResponseType) => Promise<StreamResponseReturnType>;
-  }
-> = {
-  [OpenAiChatEnum.GPT35]: {
-    chatCompletion: (data: ChatCompletionType) =>
-      chatResponse({ model: OpenAiChatEnum.GPT35, ...data }),
-    streamResponse: (data: StreamResponseType) =>
-      openAiStreamResponse({
-        model: OpenAiChatEnum.GPT35,
-        ...data
-      })
-  },
-  [OpenAiChatEnum.GPT3516k]: {
-    chatCompletion: (data: ChatCompletionType) =>
-      chatResponse({ model: OpenAiChatEnum.GPT3516k, ...data }),
-    streamResponse: (data: StreamResponseType) =>
-      openAiStreamResponse({
-        model: OpenAiChatEnum.GPT3516k,
-        ...data
-      })
-  },
-  [OpenAiChatEnum.GPT4]: {
-    chatCompletion: (data: ChatCompletionType) =>
-      chatResponse({ model: OpenAiChatEnum.GPT4, ...data }),
-    streamResponse: (data: StreamResponseType) =>
-      openAiStreamResponse({
-        model: OpenAiChatEnum.GPT4,
-        ...data
-      })
-  },
-  [OpenAiChatEnum.GPT432k]: {
-    chatCompletion: (data: ChatCompletionType) =>
-      chatResponse({ model: OpenAiChatEnum.GPT432k, ...data }),
-    streamResponse: (data: StreamResponseType) =>
-      openAiStreamResponse({
-        model: OpenAiChatEnum.GPT432k,
-        ...data
-      })
-  }
 };
 
 /* delete invalid symbol */
@@ -124,7 +85,8 @@ export const ChatContextFilter = ({
   }
 
   // 去掉 system 的 token
-  maxTokens -= modelToolMap[model].countTokens({
+  maxTokens -= modelToolMap.countTokens({
+    model,
     messages: systemPrompts
   });
 
@@ -135,7 +97,8 @@ export const ChatContextFilter = ({
   for (let i = chatPrompts.length - 1; i >= 0; i--) {
     chats.unshift(chatPrompts[i]);
 
-    const tokens = modelToolMap[model].countTokens({
+    const tokens = modelToolMap.countTokens({
+      model,
       messages: chats
     });
 
@@ -164,13 +127,14 @@ export const resStreamResponse = async ({
   res.setHeader('X-Accel-Buffering', 'no');
   res.setHeader('Cache-Control', 'no-cache, no-transform');
 
-  const { responseContent, totalTokens, finishMessages } = await modelServiceToolMap[
+  const { responseContent, totalTokens, finishMessages } = await modelServiceToolMap.streamResponse(
-    model
+    {
-  ].streamResponse({
+      chatResponse,
-    chatResponse,
+      prompts,
-    prompts,
+      res,
-    res
+      model
-  });
+    }
+  );
 
   return { responseContent, totalTokens, finishMessages };
 };
@@ -259,7 +223,8 @@ export const V2_StreamResponse = async ({
     value: responseContent
   });
 
-  const totalTokens = modelToolMap[model].countTokens({
+  const totalTokens = modelToolMap.countTokens({
+    model,
     messages: finishMessages
   });
 

client/src/service/utils/chat/openai.ts

@@ -35,7 +35,8 @@ export const chatResponse = async ({
   const adaptMessages = adaptChatItem_openAI({ messages: filterMessages, reserveId: false });
   const chatAPI = getOpenAIApi(apiKey);
 
-  const promptsToken = modelToolMap[model].countTokens({
+  const promptsToken = modelToolMap.countTokens({
+    model,
     messages: filterMessages
   });
 
@@ -116,7 +117,8 @@ export const openAiStreamResponse = async ({
       value: responseContent
     });
 
-    const totalTokens = modelToolMap[model].countTokens({
+    const totalTokens = modelToolMap.countTokens({
+      model,
       messages: finishMessages
     });
 

client/src/types/index.d.ts

@@ -21,7 +21,9 @@ declare global {
   var QRCode: any;
   var qaQueueLen: number;
   var vectorQueueLen: number;
-  var OpenAiEncMap: Record<string, Tiktoken>;
+  var OpenAiEncMap: Tiktoken;
+  var sendInformQueue: (() => Promise<void>)[];
+  var sendInformQueueLen: number;
   var systemEnv: {
     vectorMaxProcess: number;
     qaMaxProcess: number;

client/src/utils/file.ts

@@ -152,7 +152,7 @@ export const splitText_token = ({ text, maxLen }: { text: string; maxLen: number
   const slideLen = Math.floor(maxLen * 0.3);
 
   try {
-    const enc = getOpenAiEncMap()[OpenAiChatEnum.GPT35];
+    const enc = getOpenAiEncMap();
     // filter empty text. encode sentence
     const encodeText = enc.encode(text);
 

client/src/utils/plugin/index.ts

@@ -4,32 +4,8 @@ import type { ChatItemType } from '@/types/chat';
 import { countOpenAIToken, openAiSliceTextByToken } from './openai';
 import { gpt_chatItemTokenSlice } from '@/pages/api/openapi/text/gptMessagesSlice';
 
-export const modelToolMap: Record<
+export const modelToolMap = {
-  ChatModelType,
+  countTokens: countOpenAIToken,
-  {
+  sliceText: openAiSliceTextByToken,
-    countTokens: (data: { messages: ChatItemType[] }) => number;
+  tokenSlice: gpt_chatItemTokenSlice
-    sliceText: (data: { text: string; length: number }) => string;
-    tokenSlice: (data: { messages: ChatItemType[]; maxToken: number }) => ChatItemType[];
-  }
-> = {
-  [OpenAiChatEnum.GPT35]: {
-    countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT35, messages }),
-    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT35, ...data }),
-    tokenSlice: (data) => gpt_chatItemTokenSlice({ model: OpenAiChatEnum.GPT35, ...data })
-  },
-  [OpenAiChatEnum.GPT3516k]: {
-    countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT3516k, messages }),
-    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT3516k, ...data }),
-    tokenSlice: (data) => gpt_chatItemTokenSlice({ model: OpenAiChatEnum.GPT3516k, ...data })
-  },
-  [OpenAiChatEnum.GPT4]: {
-    countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT4, messages }),
-    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT4, ...data }),
-    tokenSlice: (data) => gpt_chatItemTokenSlice({ model: OpenAiChatEnum.GPT4, ...data })
-  },
-  [OpenAiChatEnum.GPT432k]: {
-    countTokens: ({ messages }) => countOpenAIToken({ model: OpenAiChatEnum.GPT432k, messages }),
-    sliceText: (data) => openAiSliceTextByToken({ model: OpenAiChatEnum.GPT432k, ...data }),
-    tokenSlice: (data) => gpt_chatItemTokenSlice({ model: OpenAiChatEnum.GPT432k, ...data })
-  }
 };

client/src/utils/plugin/openai.ts

@@ -4,7 +4,6 @@ import { ChatRoleEnum } from '@/constants/chat';
 import { ChatCompletionRequestMessageRoleEnum } from 'openai';
 import { OpenAiChatEnum } from '@/constants/model';
 import axios from 'axios';
-import dayjs from 'dayjs';
 import type { MessageItemType } from '@/pages/api/openapi/v1/chat/completions';
 
 export const getOpenAiEncMap = () => {
@@ -14,28 +13,11 @@ export const getOpenAiEncMap = () => {
   if (typeof global !== 'undefined' && global.OpenAiEncMap) {
     return global.OpenAiEncMap;
   }
-  const enc = {
+  const enc = encoding_for_model('gpt-3.5-turbo', {
-    [OpenAiChatEnum.GPT35]: encoding_for_model('gpt-3.5-turbo', {
+    '<|im_start|>': 100264,
-      '<|im_start|>': 100264,
+    '<|im_end|>': 100265,
-      '<|im_end|>': 100265,
+    '<|im_sep|>': 100266
-      '<|im_sep|>': 100266
+  });
-    }),
-    [OpenAiChatEnum.GPT3516k]: encoding_for_model('gpt-3.5-turbo', {
-      '<|im_start|>': 100264,
-      '<|im_end|>': 100265,
-      '<|im_sep|>': 100266
-    }),
-    [OpenAiChatEnum.GPT4]: encoding_for_model('gpt-4', {
-      '<|im_start|>': 100264,
-      '<|im_end|>': 100265,
-      '<|im_sep|>': 100266
-    }),
-    [OpenAiChatEnum.GPT432k]: encoding_for_model('gpt-4-32k', {
-      '<|im_start|>': 100264,
-      '<|im_end|>': 100265,
-      '<|im_sep|>': 100266
-    })
-  };
 
   if (typeof window !== 'undefined') {
     window.OpenAiEncMap = enc;
@@ -78,7 +60,7 @@ export function countOpenAIToken({
   const adaptMessages = adaptChatItem_openAI({ messages, reserveId: true });
   const token = adaptMessages.reduce((sum, item) => {
     const text = `${item.role}\n${item.content}`;
-    const enc = getOpenAiEncMap()[model];
+    const enc = getOpenAiEncMap();
     const encodeText = enc.encode(text);
     const tokens = encodeText.length + diffVal;
     return sum + tokens;
@@ -96,7 +78,7 @@ export const openAiSliceTextByToken = ({
   text: string;
   length: number;
 }) => {
-  const enc = getOpenAiEncMap()[model];
+  const enc = getOpenAiEncMap();
   const encodeText = enc.encode(text);
   const decoder = new TextDecoder();
   return decoder.decode(enc.decode(encodeText.slice(0, length)));