feat: 拆分数据自定义prompt

src/api/model.ts

@@ -49,8 +49,8 @@ export const postModelDataInput = (data: {
   data: { text: ModelDataSchema['text']; q: ModelDataSchema['q'] }[];
   data: { text: ModelDataSchema['text']; q: ModelDataSchema['q'] }[];
 }) => POST<number>(`/model/data/pushModelDataInput`, data);
 }) => POST<number>(`/model/data/pushModelDataInput`, data);
 
 
-export const postModelDataFileText = (modelId: string, text: string) =>
-  POST(`/model/data/splitData`, { modelId, text });
+export const postModelDataFileText = (data: { modelId: string; text: string; prompt: string }) =>
+  POST(`/model/data/splitData`, data);
 
 
 export const postModelDataJsonData = (
 export const postModelDataJsonData = (
   modelId: string,
   modelId: string,

src/pages/api/chat/vectorGpt.ts

@@ -118,7 +118,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
 
 
     prompts.unshift({
     prompts.unshift({
       obj: 'SYSTEM',
       obj: 'SYSTEM',
-      value: `${model.systemPrompt} 知识库内容: "${systemPrompt}"`
+      value: `${model.systemPrompt} 知识库内容是最新的，知识库内容为: "${systemPrompt}"`
     });
     });
 
 
     // 控制在 tokens 数量，防止超出
     // 控制在 tokens 数量，防止超出

src/pages/api/model/data/splitData.ts

@@ -8,8 +8,8 @@ import { encode } from 'gpt-token-utils';
 /* 拆分数据成QA */
 /* 拆分数据成QA */
 export default async function handler(req: NextApiRequest, res: NextApiResponse) {
 export default async function handler(req: NextApiRequest, res: NextApiResponse) {
   try {
   try {
-    const { text, modelId } = req.body as { text: string; modelId: string };
-    if (!text || !modelId) {
+    const { text, modelId, prompt } = req.body as { text: string; modelId: string; prompt: string };
+    if (!text || !modelId || !prompt) {
       throw new Error('参数错误');
       throw new Error('参数错误');
     }
     }
     await connectToDatabase();
     await connectToDatabase();
@@ -62,7 +62,8 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       userId,
       userId,
       modelId,
       modelId,
       rawText: text,
       rawText: text,
-      textList
+      textList,
+      prompt
     });
     });
 
 
     generateQA();
     generateQA();

src/pages/model/detail/components/SelectFileModal.tsx

@@ -8,7 +8,8 @@ import {
   ModalContent,
   ModalContent,
   ModalHeader,
   ModalHeader,
   ModalCloseButton,
   ModalCloseButton,
-  ModalBody
+  ModalBody,
+  Input
 } from '@chakra-ui/react';
 } from '@chakra-ui/react';
 import { useToast } from '@/hooks/useToast';
 import { useToast } from '@/hooks/useToast';
 import { useSelectFile } from '@/hooks/useSelectFile';
 import { useSelectFile } from '@/hooks/useSelectFile';
@@ -34,6 +35,7 @@ const SelectFileModal = ({
 }) => {
 }) => {
   const [selecting, setSelecting] = useState(false);
   const [selecting, setSelecting] = useState(false);
   const { toast } = useToast();
   const { toast } = useToast();
+  const [prompt, setPrompt] = useState('');
   const { File, onOpen } = useSelectFile({ fileType: fileExtension, multiple: true });
   const { File, onOpen } = useSelectFile({ fileType: fileExtension, multiple: true });
   const [fileText, setFileText] = useState('');
   const [fileText, setFileText] = useState('');
   const { openConfirm, ConfirmChild } = useConfirm({
   const { openConfirm, ConfirmChild } = useConfirm({
@@ -83,7 +85,11 @@ const SelectFileModal = ({
   const { mutate, isLoading } = useMutation({
   const { mutate, isLoading } = useMutation({
     mutationFn: async () => {
     mutationFn: async () => {
       if (!fileText) return;
       if (!fileText) return;
-      await postModelDataFileText(modelId, fileText);
+      await postModelDataFileText({
+        modelId,
+        text: fileText,
+        prompt: `下面是${prompt || '一段长文本'}`
+      });
       toast({
       toast({
         title: '导入数据成功,需要一段拆解和训练',
         title: '导入数据成功,需要一段拆解和训练',
         status: 'success'
         status: 'success'
@@ -102,7 +108,7 @@ const SelectFileModal = ({
   return (
   return (
     <Modal isOpen={true} onClose={onClose} isCentered>
     <Modal isOpen={true} onClose={onClose} isCentered>
       <ModalOverlay />
       <ModalOverlay />
-      <ModalContent maxW={'min(900px, 90vw)'} m={0} position={'relative'} h={['90vh', '70vh']}>
+      <ModalContent maxW={'min(900px, 90vw)'} m={0} position={'relative'} h={'90vh'}>
         <ModalHeader>文件导入</ModalHeader>
         <ModalHeader>文件导入</ModalHeader>
         <ModalCloseButton />
         <ModalCloseButton />
 
 
@@ -125,6 +131,17 @@ const SelectFileModal = ({
           <Box mt={2}>
           <Box mt={2}>
             一共 {fileText.length} 个字，{encode(fileText).length} 个tokens
             一共 {fileText.length} 个字，{encode(fileText).length} 个tokens
           </Box>
           </Box>
+          <Flex w={'100%'} alignItems={'center'} my={4}>
+            <Box flex={'0 0 auto'} mr={2}>
+              下面是
+            </Box>
+            <Input
+              placeholder="提示词，例如: Laf的介绍/关于gpt4的论文/一段长文本"
+              value={prompt}
+              onChange={(e) => setPrompt(e.target.value)}
+              size={'sm'}
+            />
+          </Flex>
           <Box
           <Box
             flex={'1 0 0'}
             flex={'1 0 0'}
             h={0}
             h={0}

src/service/events/generateQA.ts

@@ -14,11 +14,6 @@ export async function generateQA(next = false): Promise<any> {
   if (global.generatingQA && !next) return;
   if (global.generatingQA && !next) return;
   global.generatingQA = true;
   global.generatingQA = true;
 
 
-  const systemPrompt: ChatCompletionRequestMessage = {
-    role: 'system',
-    content: `总结助手。我会向你发送一段长文本,请从中总结出5至30个问题和答案,答案请尽量详细,并按以下格式返回: Q1:\nA1:\nQ2:\nA2:\n`
-  };
-
   try {
   try {
     const redis = await connectRedis();
     const redis = await connectRedis();
     // 找出一个需要生成的 dataItem
     // 找出一个需要生成的 dataItem
@@ -63,6 +58,13 @@ export async function generateQA(next = false): Promise<any> {
 
 
     // 获取 openai 请求实例
     // 获取 openai 请求实例
     const chatAPI = getOpenAIApi(userApiKey || systemKey);
     const chatAPI = getOpenAIApi(userApiKey || systemKey);
+    const systemPrompt: ChatCompletionRequestMessage = {
+      role: 'system',
+      content: `${
+        dataItem.prompt || '下面是一段长文本'
+      },请从中总结出5至30个问题和答案,答案尽量详细,并按以下格式返回: Q1:\nA1:\nQ2:\nA2:\n`
+    };
+
     // 请求 chatgpt 获取回答
     // 请求 chatgpt 获取回答
     const response = await chatAPI
     const response = await chatAPI
       .createChatCompletion(
       .createChatCompletion(

src/service/models/splitData.ts

@@ -8,6 +8,11 @@ const SplitDataSchema = new Schema({
     ref: 'user',
     ref: 'user',
     required: true
     required: true
   },
   },
+  prompt: {
+    // 拆分时的提示词
+    type: String,
+    required: true
+  },
   modelId: {
   modelId: {
     type: Schema.Types.ObjectId,
     type: Schema.Types.ObjectId,
     ref: 'model',
     ref: 'model',

src/types/mongoSchema.d.ts

@@ -69,6 +69,7 @@ export interface ModelSplitDataSchema {
   userId: string;
   userId: string;
   modelId: string;
   modelId: string;
   rawText: string;
   rawText: string;
+  prompt: string;
   errorText: string;
   errorText: string;
   textList: string[];
   textList: string[];
 }
 }