feat: gpt35-16k

client/src/constants/model.ts

@@ -7,7 +7,7 @@ export const embeddingPrice = 0.1;
 export type EmbeddingModelType = 'text-embedding-ada-002';
 
 export enum OpenAiChatEnum {
-  'GPT35' = 'gpt-3.5-turbo',
+  'GPT35' = 'gpt-3.5-turbo-16k',
   'GPT4' = 'gpt-4',
   'GPT432k' = 'gpt-4-32k'
 }
@@ -30,8 +30,8 @@ export const ChatModelMap = {
   [OpenAiChatEnum.GPT35]: {
     chatModel: OpenAiChatEnum.GPT35,
     name: 'ChatGpt',
-    contextMaxToken: 4096,
+    contextMaxToken: 16000,
-    systemMaxToken: 2700,
+    systemMaxToken: 8000,
     maxTemperature: 1.2,
     price: 2.5
   },

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

@@ -8,6 +8,7 @@ import { TrainingModeEnum } from '@/constants/plugin';
 import { startQueue } from '@/service/utils/tools';
 import { PgClient } from '@/service/pg';
 import { modelToolMap } from '@/utils/plugin';
+import { OpenAiChatEnum } from '@/constants/model';
 
 type DateItemType = { a: string; q: string; source?: string };
 
@@ -76,7 +77,7 @@ export async function pushDataToKb({
     const text = item.q + item.a;
 
     // count token
-    const token = modelToolMap['gpt-3.5-turbo'].countTokens({
+    const token = modelToolMap[OpenAiChatEnum.GPT35].countTokens({
       messages: [{ obj: 'System', value: item.q }]
     });
 

client/src/pages/api/openapi/plugin/openaiEmbedding.ts

@@ -7,6 +7,7 @@ import { embeddingModel } from '@/constants/model';
 import { axiosConfig } from '@/service/utils/tools';
 import { pushGenerateVectorBill } from '@/service/events/pushBill';
 import { ApiKeyType } from '@/service/utils/auth';
+import { OpenAiChatEnum } from '@/constants/model';
 
 type Props = {
   input: string[];
@@ -42,7 +43,7 @@ export async function openaiEmbedding({
   type = 'chat'
 }: { userId: string; mustPay?: boolean } & Props) {
   const { userOpenAiKey, systemAuthKey } = await getApiKey({
-    model: 'gpt-3.5-turbo',
+    model: OpenAiChatEnum.GPT35,
     userId,
     mustPay,
     type

client/src/pages/api/openapi/text/gptMessagesSlice.ts

@@ -4,8 +4,9 @@ import { jsonRes } from '@/service/response';
 import { authUser } from '@/service/utils/auth';
 import type { ChatItemSimpleType } from '@/types/chat';
 import { countOpenAIToken } from '@/utils/plugin/openai';
+import { OpenAiChatEnum } from '@/constants/model';
 
-type ModelType = 'gpt-3.5-turbo' | 'gpt-4' | 'gpt-4-32k';
+type ModelType = `${OpenAiChatEnum}`;
 
 type Props = {
   messages: ChatItemSimpleType[];

client/src/pages/kb/components/SelectFileModal.tsx

@@ -29,14 +29,14 @@ const fileExtension = '.txt,.doc,.docx,.pdf,.md';
 
 const modeMap = {
   [TrainingModeEnum.qa]: {
-    maxLen: 2600,
+    maxLen: 9000,
-    slideLen: 700,
+    slideLen: 3000,
     price: ChatModelMap[OpenAiChatEnum.GPT35].price,
     isPrompt: true
   },
   [TrainingModeEnum.index]: {
-    maxLen: 700,
+    maxLen: 2000,
-    slideLen: 300,
+    slideLen: 600,
     price: embeddingPrice,
     isPrompt: false
   }

client/src/service/events/generateQA.ts

@@ -96,7 +96,7 @@ export async function generateQA(): Promise<any> {
                 obj: ChatRoleEnum.System,
                 value: `你是出题人
 ${data.prompt || '下面是"一段长文本"'}
-从中选出5至20个题目和答案.答案详细.按格式返回: Q1:
+从中选出15至30个题目和答案.答案详细.按格式返回: Q1:
 A1:
 Q2:
 A2:

client/src/service/utils/auth.ts

@@ -177,7 +177,7 @@ export const getApiKey = async ({
   };
 
   // 有自己的key
-  if (!mustPay && keyMap[model].userOpenAiKey) {
+  if (!mustPay && keyMap[model]?.userOpenAiKey) {
     return {
       user,
       userOpenAiKey: keyMap[model].userOpenAiKey,

client/src/store/chat.ts

@@ -64,7 +64,7 @@ const defaultShareChatData: ShareChatType = {
     avatar: '/icon/logo.png',
     intro: ''
   },
-  chatModel: 'gpt-3.5-turbo',
+  chatModel: OpenAiChatEnum.GPT35,
   history: []
 };
 

client/src/utils/file.ts

@@ -2,6 +2,7 @@ import mammoth from 'mammoth';
 import Papa from 'papaparse';
 import { getOpenAiEncMap } from './plugin/openai';
 import { getErrText } from './tools';
+import { OpenAiChatEnum } from '@/constants/model';
 
 /**
  * 读取 txt 文件内容
@@ -156,7 +157,7 @@ export const splitText_token = ({
   slideLen: number;
 }) => {
   try {
-    const enc = getOpenAiEncMap()['gpt-3.5-turbo'];
+    const enc = getOpenAiEncMap()[OpenAiChatEnum.GPT35];
     // filter empty text. encode sentence
     const encodeText = enc.encode(text);
 

client/src/utils/plugin/openai.ts

@@ -11,17 +11,17 @@ const graphemer = new Graphemer();
 export const getOpenAiEncMap = () => {
   if (typeof window !== 'undefined') {
     window.OpenAiEncMap = window.OpenAiEncMap || {
-      'gpt-3.5-turbo': encoding_for_model('gpt-3.5-turbo', {
+      [OpenAiChatEnum.GPT35]: encoding_for_model('gpt-3.5-turbo', {
         '<|im_start|>': 100264,
         '<|im_end|>': 100265,
         '<|im_sep|>': 100266
       }),
-      'gpt-4': encoding_for_model('gpt-4', {
+      [OpenAiChatEnum.GPT4]: encoding_for_model('gpt-4', {
         '<|im_start|>': 100264,
         '<|im_end|>': 100265,
         '<|im_sep|>': 100266
       }),
-      'gpt-4-32k': encoding_for_model('gpt-4-32k', {
+      [OpenAiChatEnum.GPT432k]: encoding_for_model('gpt-4-32k', {
         '<|im_start|>': 100264,
         '<|im_end|>': 100265,
         '<|im_sep|>': 100266
@@ -31,17 +31,17 @@ export const getOpenAiEncMap = () => {
   }
   if (typeof global !== 'undefined') {
     global.OpenAiEncMap = global.OpenAiEncMap || {
-      'gpt-3.5-turbo': encoding_for_model('gpt-3.5-turbo', {
+      [OpenAiChatEnum.GPT35]: encoding_for_model('gpt-3.5-turbo', {
         '<|im_start|>': 100264,
         '<|im_end|>': 100265,
         '<|im_sep|>': 100266
       }),
-      'gpt-4': encoding_for_model('gpt-4', {
+      [OpenAiChatEnum.GPT4]: encoding_for_model('gpt-4', {
         '<|im_start|>': 100264,
         '<|im_end|>': 100265,
         '<|im_sep|>': 100266
       }),
-      'gpt-4-32k': encoding_for_model('gpt-4-32k', {
+      [OpenAiChatEnum.GPT432k]: encoding_for_model('gpt-4-32k', {
         '<|im_start|>': 100264,
         '<|im_end|>': 100265,
         '<|im_sep|>': 100266
@@ -50,17 +50,17 @@ export const getOpenAiEncMap = () => {
     return global.OpenAiEncMap;
   }
   return {
-    'gpt-3.5-turbo': encoding_for_model('gpt-3.5-turbo', {
+    [OpenAiChatEnum.GPT35]: encoding_for_model('gpt-3.5-turbo', {
       '<|im_start|>': 100264,
       '<|im_end|>': 100265,
       '<|im_sep|>': 100266
     }),
-    'gpt-4': encoding_for_model('gpt-4', {
+    [OpenAiChatEnum.GPT4]: encoding_for_model('gpt-4', {
       '<|im_start|>': 100264,
       '<|im_end|>': 100265,
       '<|im_sep|>': 100266
     }),
-    'gpt-4-32k': encoding_for_model('gpt-4-32k', {
+    [OpenAiChatEnum.GPT432k]: encoding_for_model('gpt-4-32k', {
       '<|im_start|>': 100264,
       '<|im_end|>': 100265,
       '<|im_sep|>': 100266
@@ -97,9 +97,9 @@ export function countOpenAIToken({
       content: string;
       name?: string;
     }[],
-    model: 'gpt-3.5-turbo' | 'gpt-4' | 'gpt-4-32k'
+    model: `${OpenAiChatEnum}`
   ) {
-    const isGpt3 = model === 'gpt-3.5-turbo';
+    const isGpt3 = model.startsWith('gpt-3.5-turbo');
 
     const msgSep = isGpt3 ? '\n' : '';
     const roleSep = isGpt3 ? '\n' : '<|im_sep|>';
@@ -147,7 +147,7 @@ export function countOpenAIToken({
 }
 
 export const openAiSliceTextByToken = ({
-  model = 'gpt-3.5-turbo',
+  model = OpenAiChatEnum.GPT35,
   text,
   length
 }: {