perf: model framwork

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: [

src/service/events/pushBill.ts

@@ -1,5 +1,5 @@
 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 = modelItem?.price || 3;
+        const unitPrice = ChatModelMap[OpenAiChatEnum.GPT35]?.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

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: {

src/service/tools/searchKb.ts

@@ -1,6 +1,6 @@
 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 {

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;
 }) => {