perf: model framwork

91decc36 · archer · aa74625f · 91decc36 · 91decc36 · 91decc36
Commit 91decc36 authored May 03, 2023 by archer
19 changed files
--- a/src/constants/model.ts
+++ b/src/constants/model.ts
@@ -2,33 +2,34 @@ import type { ModelSchema } from '@/types/mongoSchema';
 export const embeddingModel = 'text-embedding-ada-002';
-export enum ChatModelEnum {
+export enum OpenAiChatEnum {
  'GPT35' = 'gpt-3.5-turbo',
  'GPT4' = 'gpt-4',
  'GPT432k' = 'gpt-4-32k'
 }
-export const ChatModelMap = {
-  // ui name
-  [ChatModelEnum.GPT35]: 'ChatGpt',
-  [ChatModelEnum.GPT4]: 'Gpt4',
-  [ChatModelEnum.GPT432k]: 'Gpt4-32k'
-};
-export type ChatModelConstantType = {
+export type ChatModelType = `${OpenAiChatEnum}`;
-  chatModel: `${ChatModelEnum}`;
-  contextMaxToken: number;
-  maxTemperature: number;
-  price: number; // 多少钱 / 1token，单位: 0.00001元
-};
-export const modelList: ChatModelConstantType[] = [
+export const ChatModelMap = {
-  {
+  [OpenAiChatEnum.GPT35]: {
-    chatModel: ChatModelEnum.GPT35,
+    name: 'ChatGpt',
    contextMaxToken: 4096,
    maxTemperature: 1.5,
    price: 3
+  },
+  [OpenAiChatEnum.GPT4]: {
+    name: 'Gpt4',
+    contextMaxToken: 8000,
+    maxTemperature: 1.5,
+    price: 30
+  },
+  [OpenAiChatEnum.GPT432k]: {
+    name: 'Gpt4-32k',
+    contextMaxToken: 8000,
+    maxTemperature: 1.5,
+    price: 30
  }
-];
+};
 export enum ModelStatusEnum {
  running = 'running',
@@ -106,7 +107,7 @@ export const defaultModel: ModelSchema = {
    searchMode: ModelVectorSearchModeEnum.hightSimilarity,
    systemPrompt: '',
    temperature: 0,
-    chatModel: ChatModelEnum.GPT35
+    chatModel: OpenAiChatEnum.GPT35
  },
  share: {
    isShare: false,

--- a/src/pages/api/chat/chat.ts
+++ b/src/pages/api/chat/chat.ts
@@ -5,7 +5,7 @@ import { axiosConfig, openaiChatFilter } from '@/service/utils/tools';
 import { ChatItemSimpleType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
 import { PassThrough } from 'stream';
-import { modelList, ModelVectorSearchModeMap } from '@/constants/model';
+import { ChatModelMap, ModelVectorSearchModeMap } from '@/constants/model';
 import { pushChatBill } from '@/service/events/pushBill';
 import { gpt35StreamResponse } from '@/service/utils/openai';
 import { searchKb_openai } from '@/service/tools/searchKb';
@@ -47,10 +47,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      authorization
    });
-    const modelConstantsData = modelList.find((item) => item.chatModel === model.chat.chatModel);
+    const modelConstantsData = ChatModelMap[model.chat.chatModel];
-    if (!modelConstantsData) {
-      throw new Error('模型加载异常');
-    }
    // 读取对话内容
    const prompts = [...content, prompt];
@@ -61,7 +58,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
        apiKey: userApiKey || systemKey,
        isPay: !userApiKey,
        text: prompt.value,
-        similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity || 0.22,
+        similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
        model,
        userId
      });

--- a/src/pages/api/openapi/chat/chat.ts
+++ b/src/pages/api/openapi/chat/chat.ts
 import type { NextApiRequest, NextApiResponse } from 'next';
 import { connectToDatabase } from '@/service/mongo';
 import { getOpenAIApi, authOpenApiKey, authModel } from '@/service/utils/auth';
-import { axiosConfig, openaiChatFilter, systemPromptFilter } from '@/service/utils/tools';
+import { axiosConfig, openaiChatFilter } from '@/service/utils/tools';
 import { ChatItemSimpleType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
 import { PassThrough } from 'stream';
-import { modelList, ModelVectorSearchModeMap, ModelVectorSearchModeEnum } from '@/constants/model';
+import { ChatModelMap, ModelVectorSearchModeMap } from '@/constants/model';
 import { pushChatBill } from '@/service/events/pushBill';
 import { gpt35StreamResponse } from '@/service/utils/openai';
 import { searchKb_openai } from '@/service/tools/searchKb';
@@ -58,10 +58,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      modelId
    });
-    const modelConstantsData = modelList.find((item) => item.chatModel === model.chat.chatModel);
+    const modelConstantsData = ChatModelMap[model.chat.chatModel];
-    if (!modelConstantsData) {
-      throw new Error('模型加载异常');
-    }
    // 使用了知识库搜索
    if (model.chat.useKb) {

--- a/src/pages/api/openapi/chat/chatGpt.ts
+++ b/src/pages/api/openapi/chat/chatGpt.ts
@@ -5,7 +5,7 @@ import { axiosConfig, openaiChatFilter } from '@/service/utils/tools';
 import { ChatItemSimpleType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
 import { PassThrough } from 'stream';
-import { modelList } from '@/constants/model';
+import { ChatModelMap } from '@/constants/model';
 import { pushChatBill } from '@/service/events/pushBill';
 import { gpt35StreamResponse } from '@/service/utils/openai';
@@ -60,10 +60,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      throw new Error('无权使用该模型');
    }
-    const modelConstantsData = modelList.find((item) => item.chatModel === model.chat.chatModel);
+    const modelConstantsData = ChatModelMap[model.chat.chatModel];
-    if (!modelConstantsData) {
-      throw new Error('模型加载异常');
-    }
    // 如果有系统提示词，自动插入
    if (model.chat.systemPrompt) {

--- a/src/pages/api/openapi/chat/lafGpt.ts
+++ b/src/pages/api/openapi/chat/lafGpt.ts
 import type { NextApiRequest, NextApiResponse } from 'next';
 import { connectToDatabase, Model } from '@/service/mongo';
 import { getOpenAIApi, authOpenApiKey } from '@/service/utils/auth';
-import { axiosConfig, openaiChatFilter, systemPromptFilter } from '@/service/utils/tools';
+import { axiosConfig, openaiChatFilter } from '@/service/utils/tools';
 import { ChatItemSimpleType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
 import { PassThrough } from 'stream';
-import { modelList, ModelVectorSearchModeMap, ChatModelEnum } from '@/constants/model';
+import { ChatModelMap, ModelVectorSearchModeMap, OpenAiChatEnum } from '@/constants/model';
 import { pushChatBill } from '@/service/events/pushBill';
 import { gpt35StreamResponse } from '@/service/utils/openai';
 import { searchKb_openai } from '@/service/tools/searchKb';
@@ -53,10 +53,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      throw new Error('找不到模型');
    }
-    const modelConstantsData = modelList.find((item) => item.chatModel === model.chat.chatModel);
+    const modelConstantsData = ChatModelMap[model.chat.chatModel];
-    if (!modelConstantsData) {
-      throw new Error('model is undefined');
-    }
    console.log('laf gpt start');
@@ -66,7 +63,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
    // 请求一次 chatgpt 拆解需求
    const promptResponse = await chatAPI.createChatCompletion(
      {
-        model: ChatModelEnum.GPT35,
+        model: OpenAiChatEnum.GPT35,
        temperature: 0,
        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

--- a/src/pages/api/openapi/chat/vectorGpt.ts
+++ b/src/pages/api/openapi/chat/vectorGpt.ts
 import type { NextApiRequest, NextApiResponse } from 'next';
 import { connectToDatabase, Model } from '@/service/mongo';
-import { axiosConfig, systemPromptFilter, openaiChatFilter } from '@/service/utils/tools';
+import { axiosConfig, openaiChatFilter } from '@/service/utils/tools';
 import { getOpenAIApi, authOpenApiKey } from '@/service/utils/auth';
 import { ChatItemSimpleType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
 import { PassThrough } from 'stream';
-import { modelList, ModelVectorSearchModeMap, ModelVectorSearchModeEnum } from '@/constants/model';
+import {
+  ChatModelMap,
+  ModelVectorSearchModeMap,
+  ModelVectorSearchModeEnum
+} from '@/constants/model';
 import { pushChatBill } from '@/service/events/pushBill';
 import { gpt35StreamResponse } from '@/service/utils/openai';
 import { searchKb_openai } from '@/service/tools/searchKb';
@@ -62,10 +66,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      throw new Error('无权使用该模型');
    }
-    const modelConstantsData = modelList.find((item) => item.chatModel === model.chat.chatModel);
+    const modelConstantsData = ChatModelMap[model.chat.chatModel];
-    if (!modelConstantsData) {
-      throw new Error('模型初始化异常');
-    }
    // 获取向量匹配到的提示词
    const { code, searchPrompt } = await searchKb_openai({

--- a/src/pages/chat/index.tsx
+++ b/src/pages/chat/index.tsx
@@ -27,7 +27,7 @@ import {
 import { useToast } from '@/hooks/useToast';
 import { useScreen } from '@/hooks/useScreen';
 import { useQuery } from '@tanstack/react-query';
-import { ChatModelEnum } from '@/constants/model';
+import { OpenAiChatEnum } from '@/constants/model';
 import dynamic from 'next/dynamic';
 import { useGlobalStore } from '@/store/global';
 import { useCopyData } from '@/utils/tools';
@@ -69,7 +69,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
    name: '',
    avatar: '/icon/logo.png',
    intro: '',
-    chatModel: ChatModelEnum.GPT35,
+    chatModel: OpenAiChatEnum.GPT35,
    history: []
  }); // 聊天框整体数据

--- a/src/pages/model/detail/components/ModelEditForm.tsx
+++ b/src/pages/model/detail/components/ModelEditForm.tsx
@@ -21,7 +21,7 @@ import {
 import { QuestionOutlineIcon } from '@chakra-ui/icons';
 import type { ModelSchema } from '@/types/mongoSchema';
 import { UseFormReturn } from 'react-hook-form';
-import { ChatModelMap, modelList, ModelVectorSearchModeMap } from '@/constants/model';
+import { ChatModelMap, ModelVectorSearchModeMap } from '@/constants/model';
 import { formatPrice } from '@/utils/user';
 import { useConfirm } from '@/hooks/useConfirm';
 import { useSelectFile } from '@/hooks/useSelectFile';
@@ -110,17 +110,14 @@ const ModelEditForm = ({
          <Box flex={'0 0 80px'} w={0}>
            对话模型:
          </Box>
-          <Box>{ChatModelMap[getValues('chat.chatModel')]}</Box>
+          <Box>{ChatModelMap[getValues('chat.chatModel')].name}</Box>
        </Flex>
        <Flex alignItems={'center'} mt={5}>
          <Box flex={'0 0 80px'} w={0}>
            价格:
          </Box>
          <Box>
-            {formatPrice(
+            {formatPrice(ChatModelMap[getValues('chat.chatModel')].price, 1000)}
-              modelList.find((item) => item.chatModel === getValues('chat.chatModel'))?.price || 0,
-              1000
-            )}
            元/1K tokens(包括上下文和回答)
          </Box>
        </Flex>

--- a/src/pages/model/detail/index.tsx
+++ b/src/pages/model/detail/index.tsx
@@ -5,7 +5,7 @@ import type { ModelSchema } from '@/types/mongoSchema';
 import { Card, Box, Flex, Button, Tag, Grid } from '@chakra-ui/react';
 import { useToast } from '@/hooks/useToast';
 import { useForm } from 'react-hook-form';
-import { formatModelStatus, modelList, defaultModel } from '@/constants/model';
+import { formatModelStatus, defaultModel } from '@/constants/model';
 import { useGlobalStore } from '@/store/global';
 import { useScreen } from '@/hooks/useScreen';
 import { useQuery } from '@tanstack/react-query';

--- a/src/pages/model/list/components/ModelPhoneList.tsx
+++ b/src/pages/model/list/components/ModelPhoneList.tsx
@@ -43,7 +43,7 @@ const ModelPhoneList = ({
          </Flex>
          <Flex mt={5}>
            <Box flex={'0 0 100px'}>对话模型: </Box>
-            <Box color={'blackAlpha.500'}>{ChatModelMap[model.chat.chatModel]}</Box>
+            <Box color={'blackAlpha.500'}>{ChatModelMap[model.chat.chatModel].name}</Box>
          </Flex>
          <Flex mt={5}>
            <Box flex={'0 0 100px'}>模型温度: </Box>

--- a/src/pages/model/list/components/ModelTable.tsx
+++ b/src/pages/model/list/components/ModelTable.tsx
@@ -36,7 +36,7 @@ const ModelTable = ({
      key: 'service',
      render: (model: ModelSchema) => (
        <Box fontWeight={'bold'} whiteSpace={'pre-wrap'} maxW={'200px'}>
-          {ChatModelMap[model.chat.chatModel]}
+          {ChatModelMap[model.chat.chatModel].name}
        </Box>
      )
    },

--- a/src/pages/number/components/PayModal.tsx
+++ b/src/pages/number/components/PayModal.tsx
@@ -85,25 +85,6 @@ const PayModal = ({ onClose }: { onClose: () => void }) => {
          <ModalBody py={0}>
            {!payId && (
              <>
-                {/* 价格表 */}
-                {/* <TableContainer mb={4}>
-                  <Table>
-                    <Thead>
-                      <Tr>
-                        <Th>模型类型</Th>
-                        <Th>价格(元/1K tokens，包含所有上下文)</Th>
-                      </Tr>
-                    </Thead>
-                    <Tbody>
-                      {modelList.map((item, i) => (
-                        <Tr key={item.model}>
-                          <Td>{item.name}</Td>
-                          <Td>{formatPrice(item.price, 1000)}</Td>
-                        </Tr>
-                      ))}
-                    </Tbody>
-                  </Table>
-                </TableContainer> */}
                <Grid gridTemplateColumns={'repeat(4,1fr)'} gridGap={5} mb={4}>
                  {[5, 10, 20, 50].map((item) => (
                    <Button

--- a/src/service/events/generateQA.ts
+++ b/src/service/events/generateQA.ts
@@ -3,7 +3,7 @@ import { getOpenAIApi } from '@/service/utils/auth';
 import { axiosConfig } from '@/service/utils/tools';
 import { getOpenApiKey } from '../utils/openai';
 import type { ChatCompletionRequestMessage } from 'openai';
-import { ChatModelEnum } from '@/constants/model';
+import { OpenAiChatEnum } from '@/constants/model';
 import { pushSplitDataBill } from '@/service/events/pushBill';
 import { generateVector } from './generateVector';
 import { openaiError2 } from '../errorCode';
@@ -88,7 +88,7 @@ A2:
        chatAPI
          .createChatCompletion(
            {
-              model: ChatModelEnum.GPT35,
+              model: OpenAiChatEnum.GPT35,
              temperature: 0.8,
              n: 1,
              messages: [

--- a/src/service/events/pushBill.ts
+++ b/src/service/events/pushBill.ts
 import { connectToDatabase, Bill, User } from '../mongo';
-import { modelList, ChatModelEnum, embeddingModel } from '@/constants/model';
+import { ChatModelMap, OpenAiChatEnum, ChatModelType, embeddingModel } from '@/constants/model';
 import { BillTypeEnum } from '@/constants/user';
 import { countChatTokens } from '@/utils/tools';
@@ -11,7 +11,7 @@ export const pushChatBill = async ({
  messages
 }: {
  isPay: boolean;
-  chatModel: `${ChatModelEnum}`;
+  chatModel: ChatModelType;
  userId: string;
  chatId?: '' | string;
  messages: { role: 'system' | 'user' | 'assistant'; content: string }[];
@@ -30,10 +30,8 @@ export const pushChatBill = async ({
    if (isPay) {
      await connectToDatabase();
-      // 获取模型单价格
-      const modelItem = modelList.find((item) => item.chatModel === chatModel);
      // 计算价格
-      const unitPrice = modelItem?.price || 5;
+      const unitPrice = ChatModelMap[chatModel]?.price || 5;
      const price = unitPrice * tokens;
      try {
@@ -88,8 +86,7 @@ export const pushSplitDataBill = async ({
    if (isPay) {
      try {
        // 获取模型单价格, 都是用 gpt35 拆分
-        const modelItem = modelList.find((item) => item.chatModel === ChatModelEnum.GPT35);
+        const unitPrice = ChatModelMap[OpenAiChatEnum.GPT35]?.price || 3;
-        const unitPrice = modelItem?.price || 3;
        // 计算价格
        const price = unitPrice * tokenLen;
@@ -97,7 +94,7 @@ export const pushSplitDataBill = async ({
        const res = await Bill.create({
          userId,
          type,
-          modelName: ChatModelEnum.GPT35,
+          modelName: OpenAiChatEnum.GPT35,
          textLen: text.length,
          tokenLen,
          price

--- a/src/service/models/model.ts
+++ b/src/service/models/model.ts
@@ -4,7 +4,7 @@ import {
  ModelVectorSearchModeMap,
  ModelVectorSearchModeEnum,
  ChatModelMap,
-  ChatModelEnum
+  OpenAiChatEnum
 } from '@/constants/model';
 const ModelSchema = new Schema({
@@ -57,7 +57,7 @@ const ModelSchema = new Schema({
      // 聊天时使用的模型
      type: String,
      enum: Object.keys(ChatModelMap),
-      default: ChatModelEnum.GPT35
+      default: OpenAiChatEnum.GPT35
    }
  },
  share: {

--- a/src/service/tools/searchKb.ts
+++ b/src/service/tools/searchKb.ts
 import { openaiCreateEmbedding } from '../utils/openai';
 import { PgClient } from '@/service/pg';
-import { ModelDataStatusEnum, ModelVectorSearchModeEnum } from '@/constants/model';
+import { ModelDataStatusEnum, ModelVectorSearchModeEnum, ChatModelMap } from '@/constants/model';
 import { ModelSchema } from '@/types/mongoSchema';
 import { systemPromptFilter } from '../utils/tools';
@@ -9,9 +9,9 @@ import { systemPromptFilter } from '../utils/tools';
 */
 export const searchKb_openai = async ({
  apiKey,
-  isPay,
+  isPay = true,
  text,
-  similarity,
+  similarity = 0.2,
  model,
  userId
 }: {
@@ -20,7 +20,7 @@ export const searchKb_openai = async ({
  text: string;
  model: ModelSchema;
  userId: string;
-  similarity: number;
+  similarity?: number;
 }): Promise<{
  code: 200 | 201;
  searchPrompt?: {
@@ -28,6 +28,8 @@ export const searchKb_openai = async ({
    value: string;
  };
 }> => {
+  const modelConstantsData = ChatModelMap[model.chat.chatModel];
  // 获取提示词的向量
  const { vector: promptVector } = await openaiCreateEmbedding({
    isPay,
@@ -78,11 +80,11 @@ export const searchKb_openai = async ({
  }
  // 有匹配情况下，system 添加知识库内容。
-  // 系统提示词过滤，最多 2500 tokens
+  // 系统提示词过滤，最多 65% tokens
  const filterSystemPrompt = systemPromptFilter({
    model: model.chat.chatModel,
    prompts: systemPrompts,
-    maxTokens: 2500
+    maxTokens: Math.floor(modelConstantsData.contextMaxToken * 0.65)
  });
  return {

--- a/src/service/utils/tools.ts
+++ b/src/service/utils/tools.ts
@@ -3,7 +3,7 @@ import jwt from 'jsonwebtoken';
 import { ChatItemSimpleType } from '@/types/chat';
 import { countChatTokens, sliceTextByToken } from '@/utils/tools';
 import { ChatCompletionRequestMessageRoleEnum, ChatCompletionRequestMessage } from 'openai';
-import { ChatModelEnum } from '@/constants/model';
+import type { ChatModelType } from '@/constants/model';
 /* 密码加密 */
 export const hashPassword = (psw: string) => {
@@ -44,7 +44,7 @@ export const openaiChatFilter = ({
  prompts,
  maxTokens
 }: {
-  model: `${ChatModelEnum}`;
+  model: ChatModelType;
  prompts: ChatItemSimpleType[];
  maxTokens: number;
 }) => {

--- a/src/types/mongoSchema.d.ts
+++ b/src/types/mongoSchema.d.ts
@@ -3,7 +3,7 @@ import {
  ModelStatusEnum,
  ModelNameEnum,
  ModelVectorSearchModeEnum,
-  ChatModelEnum
+  ChatModelType
 } from '@/constants/model';
 import type { DataType } from './data';
@@ -41,7 +41,7 @@ export interface ModelSchema {
    searchMode: `${ModelVectorSearchModeEnum}`;
    systemPrompt: string;
    temperature: number;
-    chatModel: `${ChatModelEnum}`; // 聊天时用的模型，训练后就是训练的模型
+    chatModel: ChatModelType; // 聊天时用的模型，训练后就是训练的模型
  };
  share: {
    isShare: boolean;

--- a/src/utils/tools.ts
+++ b/src/utils/tools.ts
@@ -2,7 +2,7 @@ import crypto from 'crypto';
 import { useToast } from '@/hooks/useToast';
 import { encoding_for_model, type Tiktoken } from '@dqbd/tiktoken';
 import Graphemer from 'graphemer';
-import { ChatModelEnum } from '@/constants/model';
+import type { ChatModelType } from '@/constants/model';
 const textDecoder = new TextDecoder();
 const graphemer = new Graphemer();
@@ -130,7 +130,7 @@ export const countChatTokens = ({
  model = 'gpt-3.5-turbo',
  messages
 }: {
-  model?: `${ChatModelEnum}`;
+  model?: ChatModelType;
  messages: { role: 'system' | 'user' | 'assistant'; content: string }[];
 }) => {
  const text = getChatGPTEncodingText(messages, model);
@@ -142,7 +142,7 @@ export const sliceTextByToken = ({
  text,
  length
 }: {
-  model?: `${ChatModelEnum}`;
+  model?: ChatModelType;
  text: string;
  length: number;
 }) => {