fix: btn位置

.dockerignore

@@ -8,3 +8,4 @@ README.md
 
 .yalc/
 yalc.lock
+testApi/

Makefile

@@ -34,7 +34,7 @@ run: ## Run a dev service from host.
 
 .PHONY: docker-build
 docker-build: ## Build docker image with the desktop-frontend.
-	docker build -t c121914yu/fast-gpt:latest .
+	docker build -t c121914yu/fast-gpt:latest . --network host  --build-arg HTTP_PROXY=http://127.0.0.1:7890 --build-arg HTTPS_PROXY=http://127.0.0.1:7890
 
 ##@ Deployment
 

public/docs/csvSelect.md

@@ -1,5 +1,5 @@
 接受一个csv文件，表格头包含 question 和 answer。question 代表问题，answer 代表答案。 
-导入前会进行去重，如果问题和答案完全相同，则不会被导入，所以最终导入的内容可能会比文件的内容少。  
+导入前会进行去重，如果问题和答案完全相同，则不会被导入，所以最终导入的内容可能会比文件的内容少。但是，对于带有换行的内容，目前无法去重。
 | question | answer |
 | --- | --- | 
 | 什么是 laf  | laf 是一个云函数开发平台…… |

src/api/model.ts

@@ -49,6 +49,7 @@ export const getModelTrainings = (id: string) =>
 
 type GetModelDataListProps = RequestPaging & {
   modelId: string;
+  searchText: string;
 };
 /**
  * 获取模型的知识库数据

src/components/Markdown/index.module.scss

@@ -160,7 +160,7 @@
 }
 .markdown ul,
 .markdown ol {
-  padding-left: 1em;
+  padding-left: 2em;
 }
 .markdown ul.no-list,
 .markdown ol.no-list {

src/constants/model.ts

@@ -4,14 +4,12 @@ import type { RedisModelDataItemType } from '@/types/redis';
 export enum ChatModelNameEnum {
   GPT35 = 'gpt-3.5-turbo',
   VECTOR_GPT = 'VECTOR_GPT',
-  GPT3 = 'text-davinci-003',
   VECTOR = 'text-embedding-ada-002'
 }
 
 export const ChatModelNameMap = {
   [ChatModelNameEnum.GPT35]: 'gpt-3.5-turbo',
   [ChatModelNameEnum.VECTOR_GPT]: 'gpt-3.5-turbo',
-  [ChatModelNameEnum.GPT3]: 'text-davinci-003',
   [ChatModelNameEnum.VECTOR]: 'text-embedding-ada-002'
 };
 
@@ -34,7 +32,7 @@ export const modelList: ModelConstantsData[] = [
     trainName: '',
     maxToken: 4000,
     contextMaxToken: 7500,
-    maxTemperature: 2,
+    maxTemperature: 1.5,
     price: 3
   },
   {
@@ -47,16 +45,6 @@ export const modelList: ModelConstantsData[] = [
     maxTemperature: 1,
     price: 3
   }
-  // {
-  //   serviceCompany: 'openai',
-  //   name: 'GPT3',
-  //   model: ChatModelNameEnum.GPT3,
-  //   trainName: 'davinci',
-  //   maxToken: 4000,
-  //   contextMaxToken: 7500,
-  //   maxTemperature: 2,
-  //   price: 30
-  // }
 ];
 
 export enum TrainingStatusEnum {
@@ -97,6 +85,34 @@ export const ModelDataStatusMap: Record<RedisModelDataItemType['status'], string
   waiting: '训练中'
 };
 
+/* 知识库搜索时的配置 */
+// 搜索方式
+export enum ModelVectorSearchModeEnum {
+  hightSimilarity = 'hightSimilarity', // 高相似度+禁止回复
+  lowSimilarity = 'lowSimilarity', // 低相似度
+  noContext = 'noContex' // 高相似度+无上下文回复
+}
+export const ModelVectorSearchModeMap: Record<
+  `${ModelVectorSearchModeEnum}`,
+  {
+    text: string;
+    similarity: number;
+  }
+> = {
+  [ModelVectorSearchModeEnum.hightSimilarity]: {
+    text: '高相似度, 无匹配时拒绝回复',
+    similarity: 0.2
+  },
+  [ModelVectorSearchModeEnum.noContext]: {
+    text: '高相似度，无匹配时直接回复',
+    similarity: 0.2
+  },
+  [ModelVectorSearchModeEnum.lowSimilarity]: {
+    text: '低相似度匹配',
+    similarity: 0.8
+  }
+};
+
 export const defaultModel: ModelSchema = {
   _id: '',
   userId: '',
@@ -108,6 +124,9 @@ export const defaultModel: ModelSchema = {
   systemPrompt: '',
   intro: '',
   temperature: 5,
+  search: {
+    mode: ModelVectorSearchModeEnum.hightSimilarity
+  },
   service: {
     company: 'openai',
     trainId: '',

src/pages/api/chat/vectorGpt.ts

@@ -7,12 +7,13 @@ import { ChatItemType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
 import type { ModelSchema } from '@/types/mongoSchema';
 import { PassThrough } from 'stream';
-import { modelList } from '@/constants/model';
+import { modelList, ModelVectorSearchModeMap, ModelVectorSearchModeEnum } from '@/constants/model';
 import { pushChatBill } from '@/service/events/pushBill';
 import { connectRedis } from '@/service/redis';
 import { VecModelDataPrefix } from '@/constants/redis';
 import { vectorToBuffer } from '@/utils/tools';
 import { openaiCreateEmbedding, gpt35StreamResponse } from '@/service/utils/openai';
+import dayjs from 'dayjs';
 
 /* 发送提示词 */
 export default async function handler(req: NextApiRequest, res: NextApiResponse) {
@@ -64,13 +65,14 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       text: prompt.value
     });
 
+    const similarity = ModelVectorSearchModeMap[model.search.mode]?.similarity || 0.22;
     // 搜索系统提示词, 按相似度从 redis 中搜出相关的 q 和 text
     const redisData: any[] = await redis.sendCommand([
       'FT.SEARCH',
       `idx:${VecModelDataPrefix}:hash`,
       `@modelId:{${String(
         chat.modelId._id
-      )}} @vector:[VECTOR_RANGE 0.24 $blob]=>{$YIELD_DISTANCE_AS: score}`,
+      )}} @vector:[VECTOR_RANGE ${similarity} $blob]=>{$YIELD_DISTANCE_AS: score}`,
       'RETURN',
       '1',
       'text',
@@ -96,17 +98,34 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       }
     }
 
-    if (formatRedisPrompt.length === 0) {
-      throw new Error('对不起，我没有找到你的问题');
+    /* 高相似度+退出，无法匹配时直接退出 */
+    if (
+      formatRedisPrompt.length === 0 &&
+      model.search.mode === ModelVectorSearchModeEnum.hightSimilarity
+    ) {
+      return res.send('对不起，你的问题不在知识库中。');
     }
+    /* 高相似度+无上下文，不添加额外知识 */
+    if (
+      formatRedisPrompt.length === 0 &&
+      model.search.mode === ModelVectorSearchModeEnum.noContext
+    ) {
+      prompts.unshift({
+        obj: 'SYSTEM',
+        value: model.systemPrompt
+      });
+    } else {
+      // 有匹配情况下，添加知识库内容。
+      // 系统提示词过滤，最多 2800 tokens
+      const systemPrompt = systemPromptFilter(formatRedisPrompt, 2800);
 
-    // textArr 筛选，最多 2800 tokens
-    const systemPrompt = systemPromptFilter(formatRedisPrompt, 2800);
-
-    prompts.unshift({
-      obj: 'SYSTEM',
-      value: `${model.systemPrompt} 知识库内容是最新的,知识库内容为: "${systemPrompt}"`
-    });
+      prompts.unshift({
+        obj: 'SYSTEM',
+        value: `${model.systemPrompt} 用知识库内容回答，知识库内容为: "当前时间:${dayjs().format(
+          'YYYY/MM/DD HH:mm:ss'
+        )} ${systemPrompt}"`
+      });
+    }
 
     // 控制在 tokens 数量，防止超出
     const filterPrompts = openaiChatFilter(prompts, modelConstantsData.contextMaxToken);

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

@@ -4,7 +4,6 @@ import { connectToDatabase } from '@/service/mongo';
 import { authToken } from '@/service/utils/tools';
 import { connectRedis } from '@/service/redis';
 import { VecModelDataIdx } from '@/constants/redis';
-import { clearStrLineBreak } from '@/utils/tools';
 
 export default async function handler(req: NextApiRequest, res: NextApiResponse<any>) {
   try {
@@ -45,7 +44,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse<
 
     searchRes.documents.forEach((item: any) => {
       if (item.value.q && item.value.text) {
-        data.push([clearStrLineBreak(item.value.q), clearStrLineBreak(item.value.text)]);
+        data.push([item.value.q.replace(/\n/g, '\\n'), item.value.text.replace(/\n/g, '\\n')]);
       }
     });
 

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

@@ -4,20 +4,20 @@ import { connectToDatabase } from '@/service/mongo';
 import { authToken } from '@/service/utils/tools';
 import { connectRedis } from '@/service/redis';
 import { VecModelDataIdx } from '@/constants/redis';
-import { SearchOptions } from 'redis';
 
 export default async function handler(req: NextApiRequest, res: NextApiResponse<any>) {
   try {
     let {
       modelId,
       pageNum = 1,
-      pageSize = 10
+      pageSize = 10,
+      searchText = ''
     } = req.query as {
       modelId: string;
       pageNum: string;
       pageSize: string;
+      searchText: string;
     };
-
     const { authorization } = req.headers;
 
     pageNum = +pageNum;
@@ -40,7 +40,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse<
     // 从 redis 中获取数据
     const searchRes = await redis.ft.search(
       VecModelDataIdx,
-      `@modelId:{${modelId}} @userId:{${userId}}`,
+      `@modelId:{${modelId}} @userId:{${userId}} ${searchText ? `*${searchText}*` : ''}`,
       {
         RETURN: ['q', 'text', 'status'],
         LIMIT: {

src/pages/api/model/update.ts

@@ -8,7 +8,7 @@ import type { ModelUpdateParams } from '@/types/model';
 /* 获取我的模型 */
 export default async function handler(req: NextApiRequest, res: NextApiResponse<any>) {
   try {
-    const { name, service, security, systemPrompt, intro, temperature } =
+    const { name, search, service, security, systemPrompt, intro, temperature } =
       req.body as ModelUpdateParams;
     const { modelId } = req.query as { modelId: string };
     const { authorization } = req.headers;
@@ -37,6 +37,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse<
         systemPrompt,
         intro,
         temperature,
+        search,
         // service,
         security
       }

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

@@ -0,0 +1,158 @@
+import type { NextApiRequest, NextApiResponse } from 'next';
+import { connectToDatabase, Model } from '@/service/mongo';
+import { getOpenAIApi } from '@/service/utils/chat';
+import { httpsAgent, openaiChatFilter, authOpenApiKey } from '@/service/utils/tools';
+import { ChatCompletionRequestMessage, ChatCompletionRequestMessageRoleEnum } from 'openai';
+import { ChatItemType } from '@/types/chat';
+import { jsonRes } from '@/service/response';
+import { PassThrough } from 'stream';
+import { modelList } from '@/constants/model';
+import { pushChatBill } from '@/service/events/pushBill';
+import { gpt35StreamResponse } from '@/service/utils/openai';
+
+/* 发送提示词 */
+export default async function handler(req: NextApiRequest, res: NextApiResponse) {
+  let step = 0; // step=1时，表示开始了流响应
+  const stream = new PassThrough();
+  stream.on('error', () => {
+    console.log('error: ', 'stream error');
+    stream.destroy();
+  });
+  res.on('close', () => {
+    stream.destroy();
+  });
+  res.on('error', () => {
+    console.log('error: ', 'request error');
+    stream.destroy();
+  });
+
+  try {
+    const {
+      prompts,
+      modelId,
+      isStream = true
+    } = req.body as {
+      prompts: ChatItemType[];
+      modelId: string;
+      isStream: boolean;
+    };
+
+    if (!prompts || !modelId) {
+      throw new Error('缺少参数');
+    }
+    if (!Array.isArray(prompts)) {
+      throw new Error('prompts is not array');
+    }
+    if (prompts.length > 30 || prompts.length === 0) {
+      throw new Error('prompts length range 1-30');
+    }
+
+    await connectToDatabase();
+    let startTime = Date.now();
+
+    const { apiKey, userId } = await authOpenApiKey(req);
+
+    const model = await Model.findOne({
+      _id: modelId,
+      userId
+    });
+
+    if (!model) {
+      throw new Error('无权使用该模型');
+    }
+
+    const modelConstantsData = modelList.find((item) => item.model === model.service.modelName);
+    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.systemPrompt) {
+      prompts.unshift({
+        obj: 'SYSTEM',
+        value: model.systemPrompt
+      });
+    }
+
+    // 控制在 tokens 数量，防止超出
+    const filterPrompts = openaiChatFilter(prompts, modelConstantsData.contextMaxToken);
+
+    // 格式化文本内容成 chatgpt 格式
+    const map = {
+      Human: ChatCompletionRequestMessageRoleEnum.User,
+      AI: ChatCompletionRequestMessageRoleEnum.Assistant,
+      SYSTEM: ChatCompletionRequestMessageRoleEnum.System
+    };
+    const formatPrompts: ChatCompletionRequestMessage[] = filterPrompts.map(
+      (item: ChatItemType) => ({
+        role: map[item.obj],
+        content: item.value
+      })
+    );
+    // console.log(formatPrompts);
+    // 计算温度
+    const temperature = modelConstantsData.maxTemperature * (model.temperature / 10);
+
+    // 获取 chatAPI
+    const chatAPI = getOpenAIApi(apiKey);
+    // 发出请求
+    const chatResponse = await chatAPI.createChatCompletion(
+      {
+        model: model.service.chatModel,
+        temperature: temperature,
+        // max_tokens: modelConstantsData.maxToken,
+        messages: formatPrompts,
+        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
+        stream: isStream,
+        stop: ['.!?。']
+      },
+      {
+        timeout: 40000,
+        responseType: isStream ? 'stream' : 'json',
+        httpsAgent: httpsAgent(true)
+      }
+    );
+
+    console.log('api response time:', `${(Date.now() - startTime) / 1000}s`);
+
+    step = 1;
+    let responseContent = '';
+
+    if (isStream) {
+      const streamResponse = await gpt35StreamResponse({
+        res,
+        stream,
+        chatResponse
+      });
+      responseContent = streamResponse.responseContent;
+    } else {
+      responseContent = chatResponse.data.choices?.[0]?.message?.content || '';
+      jsonRes(res, {
+        data: responseContent
+      });
+    }
+
+    const promptsContent = formatPrompts.map((item) => item.content).join('');
+
+    // 只有使用平台的 key 才计费
+    pushChatBill({
+      isPay: true,
+      modelName: model.service.modelName,
+      userId,
+      text: promptsContent + responseContent
+    });
+  } catch (err: any) {
+    if (step === 1) {
+      // 直接结束流
+      console.log('error，结束');
+      stream.destroy();
+    } else {
+      res.status(500);
+      jsonRes(res, {
+        code: 500,
+        error: err
+      });
+    }
+  }
+}

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

@@ -83,25 +83,26 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
 下面是一些例子:
 实现一个手机号发生注册验证码方法.
 1. 从 query 中获取 phone.
-2. 校验手机号格式是否正确,不正确返回{error: "手机号格式错误"}.
+2. 校验手机号格式是否正确,不正确则返回错误码501,原因为:手机号格式错误.
 3. 给 phone 发送一个短信验证码,验证码长度为6位字符串,内容为:你正在注册laf,验证码为:code.
 4. 数据库添加数据,表为"codes",内容为 {phone, code}.
 
 实现根据手机号注册账号,需要验证手机验证码.
 1. 从 body 中获取 phone 和 code.
-2. 校验手机号格式是否正确,不正确返回{error: "手机号格式错误"}.
-2. 获取数据库数据,表为"codes",查找是否有符合 phone, code 等于body参数的记录,没有的话返回 {error:"验证码不正确"}.
+2. 校验手机号格式是否正确,不正确则返回错误码501,原因为:手机号格式错误.
+2. 获取数据库数据,表为"codes",查找是否有符合 phone, code 等于body参数的记录,没有的话返回错误码500,原因为:验证码不正确.
 4. 添加数据库数据,表为"users" ,内容为{phone, code, createTime}.
 5. 删除数据库数据,删除 code 记录.
+6. 返回新建用户的Id: return {userId}
 
 更新博客记录。传入blogId,blogText,tags,还需要记录更新的时间.
 1. 从 body 中获取 blogId,blogText 和 tags.
-2. 校验 blogId 是否为空,为空则返回 {error: "博客ID不能为空"}.
-3. 校验 blogText 是否为空,为空则返回 {error: "博客内容不能为空"}.
-4. 校验 tags 是否为数组,不是则返回 {error: "标签必须为数组"}.
+2. 校验 blogId 是否为空,为空则返回错误码500,原因为:博客ID不能为空.
+3. 校验 blogText 是否为空,为空则返回错误码500,原因为:博客内容不能为空.
+4. 校验 tags 是否为数组,不是则返回错误码500,原因为:标签必须为数组.
 5. 获取当前时间,记录为 updateTime.
 6. 更新数据库数据,表为"blogs",更新符合 blogId 的记录的内容为{blogText, tags, updateTime}.
-7. 返回结果 {message: "更新博客记录成功"}.`
+7. 返回结果 "更新博客记录成功"`
           },
           {
             role: 'user',
@@ -161,8 +162,8 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       }
     }
 
-    // textArr 筛选，最多 3200 tokens
-    const systemPrompt = systemPromptFilter(formatRedisPrompt, 3200);
+    // textArr 筛选，最多 3000 tokens
+    const systemPrompt = systemPromptFilter(formatRedisPrompt, 3000);
 
     prompts.unshift({
       obj: 'SYSTEM',

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

@@ -10,12 +10,13 @@ import { ChatCompletionRequestMessage, ChatCompletionRequestMessageRoleEnum } fr
 import { ChatItemType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
 import { PassThrough } from 'stream';
-import { modelList } from '@/constants/model';
+import { modelList, ModelVectorSearchModeMap, ModelVectorSearchModeEnum } from '@/constants/model';
 import { pushChatBill } from '@/service/events/pushBill';
 import { connectRedis } from '@/service/redis';
 import { VecModelDataPrefix } from '@/constants/redis';
 import { vectorToBuffer } from '@/utils/tools';
 import { openaiCreateEmbedding, gpt35StreamResponse } from '@/service/utils/openai';
+import dayjs from 'dayjs';
 
 /* 发送提示词 */
 export default async function handler(req: NextApiRequest, res: NextApiResponse) {
@@ -83,11 +84,13 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       text: prompts[prompts.length - 1].value // 取最后一个
     });
 
+    // 搜索系统提示词, 按相似度从 redis 中搜出相关的 q 和 text
+    const similarity = ModelVectorSearchModeMap[model.search.mode]?.similarity || 0.22;
     // 搜索系统提示词, 按相似度从 redis 中搜出相关的 q 和 text
     const redisData: any[] = await redis.sendCommand([
       'FT.SEARCH',
       `idx:${VecModelDataPrefix}:hash`,
-      `@modelId:{${modelId}} @vector:[VECTOR_RANGE 0.24 $blob]=>{$YIELD_DISTANCE_AS: score}`,
+      `@modelId:{${modelId}} @vector:[VECTOR_RANGE ${similarity} $blob]=>{$YIELD_DISTANCE_AS: score}`,
       'RETURN',
       '1',
       'text',
@@ -114,22 +117,39 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       }
     }
 
-    if (formatRedisPrompt.length === 0) {
-      throw new Error('对不起，我没有找到你的问题');
-    }
-
     // system 合并
     if (prompts[0].obj === 'SYSTEM') {
       formatRedisPrompt.unshift(prompts.shift()?.value || '');
     }
 
-    // 系统提示词筛选，最多 2800 tokens
-    const systemPrompt = systemPromptFilter(formatRedisPrompt, 2800);
+    /* 高相似度+退出，无法匹配时直接退出 */
+    if (
+      formatRedisPrompt.length === 0 &&
+      model.search.mode === ModelVectorSearchModeEnum.hightSimilarity
+    ) {
+      return res.send('对不起，你的问题不在知识库中。');
+    }
+    /* 高相似度+无上下文，不添加额外知识 */
+    if (
+      formatRedisPrompt.length === 0 &&
+      model.search.mode === ModelVectorSearchModeEnum.noContext
+    ) {
+      prompts.unshift({
+        obj: 'SYSTEM',
+        value: model.systemPrompt
+      });
+    } else {
+      // 有匹配或者低匹配度模式情况下，添加知识库内容。
+      // 系统提示词过滤，最多 2800 tokens
+      const systemPrompt = systemPromptFilter(formatRedisPrompt, 2800);
 
-    prompts.unshift({
-      obj: 'SYSTEM',
-      value: `${model.systemPrompt} 知识库内容是最新的,知识库内容为: "${systemPrompt}"`
-    });
+      prompts.unshift({
+        obj: 'SYSTEM',
+        value: `${model.systemPrompt} 用知识库内容回答，知识库内容为: "当前时间:${dayjs().format(
+          'YYYY/MM/DD HH:mm:ss'
+        )} ${systemPrompt}"`
+      });
+    }
 
     // 控制在 tokens 数量，防止超出
     const filterPrompts = openaiChatFilter(prompts, modelConstantsData.contextMaxToken);

src/pages/chat/components/SlideBar.tsx

@@ -35,13 +35,11 @@ import WxConcat from '@/components/WxConcat';
 import { useMarkdown } from '@/hooks/useMarkdown';
 
 const SlideBar = ({
-  name,
   chatId,
   modelId,
   resetChat,
   onClose
 }: {
-  name?: string;
   chatId: string;
   modelId: string;
   resetChat: () => void;
@@ -188,14 +186,14 @@ const SlideBar = ({
                     }}
                     fontSize={'xs'}
                     border={'1px solid transparent'}
-                    {...(item.name === name
+                    {...(item._id === modelId
                       ? {
                           borderColor: 'rgba(255,255,255,0.5)',
                           backgroundColor: 'rgba(255,255,255,0.1)'
                         }
                       : {})}
                     onClick={async () => {
-                      if (item.name === name) return;
+                      if (item._id === modelId) return;
                       router.replace(`/chat?chatId=${await getChatSiteId(item._id)}`);
                       onClose();
                     }}

src/pages/chat/index.tsx

@@ -114,8 +114,7 @@ const Chat = ({ chatId }: { chatId: string }) => {
     async (prompts: ChatSiteItemType) => {
       const urlMap: Record<string, string> = {
         [ChatModelNameEnum.GPT35]: '/api/chat/chatGpt',
-        [ChatModelNameEnum.VECTOR_GPT]: '/api/chat/vectorGpt',
-        [ChatModelNameEnum.GPT3]: '/api/chat/gpt3'
+        [ChatModelNameEnum.VECTOR_GPT]: '/api/chat/vectorGpt'
       };
 
       if (!urlMap[chatData.modelName]) return Promise.reject('找不到模型');
@@ -362,7 +361,6 @@ const Chat = ({ chatId }: { chatId: string }) => {
         <Box flex={'0 0 250px'} w={0} h={'100%'}>
           <SlideBar
             resetChat={resetChat}
-            name={chatData?.name}
             chatId={chatId}
             modelId={chatData.modelId}
             onClose={onCloseSlider}
@@ -394,7 +392,6 @@ const Chat = ({ chatId }: { chatId: string }) => {
             <DrawerContent maxWidth={'250px'}>
               <SlideBar
                 resetChat={resetChat}
-                name={chatData?.name}
                 chatId={chatId}
                 modelId={chatData.modelId}
                 onClose={onCloseSlider}

src/pages/data/components/ImportDataModal.tsx

@@ -15,7 +15,7 @@ import {
 import { useTabs } from '@/hooks/useTabs';
 import { useConfirm } from '@/hooks/useConfirm';
 import { useSelectFile } from '@/hooks/useSelectFile';
-import { readTxtContent, readPdfContent, readDocContent } from '@/utils/tools';
+import { readTxtContent, readPdfContent, readDocContent } from '@/utils/file';
 import { postSplitData } from '@/api/data';
 import { useMutation } from '@tanstack/react-query';
 import { useToast } from '@/hooks/useToast';

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

@@ -124,7 +124,9 @@ const InputDataModal = ({
           <Box flex={2} mr={[0, 4]} mb={[4, 0]} h={['230px', '100%']}>
             <Box h={'30px'}>问题</Box>
             <Textarea
-              placeholder="相关问题，可以回车输入多个问法, 最多500字"
+              placeholder={
+                '相关问题，可以输入多个问法, 最多500字。例如：\n1. laf 是什么？\n2. laf 可以做什么？\n3. laf怎么用'
+              }
               maxLength={500}
               resize={'none'}
               h={'calc(100% - 30px)'}
@@ -136,7 +138,9 @@ const InputDataModal = ({
           <Box flex={3} h={['330px', '100%']}>
             <Box h={'30px'}>知识点</Box>
             <Textarea
-              placeholder="知识点,最多1000字"
+              placeholder={
+                '知识点，最多1000字。请保持主语的完整性，缺少主语会导致效果不佳。例如：\n1. laf是一个云函数开发平台。\n2. laf 什么都能做。\n3. 下面是使用 laf 的例子: ……'
+              }
               maxLength={1000}
               resize={'none'}
               h={'calc(100% - 30px)'}

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

@@ -15,7 +15,8 @@ import {
   Menu,
   MenuButton,
   MenuList,
-  MenuItem
+  MenuItem,
+  Input
 } from '@chakra-ui/react';
 import type { ModelSchema } from '@/types/mongoSchema';
 import type { RedisModelDataItemType } from '@/types/redis';
@@ -40,9 +41,11 @@ const SelectFileModel = dynamic(() => import('./SelectFileModal'));
 const SelectUrlModel = dynamic(() => import('./SelectUrlModal'));
 const SelectCsvModal = dynamic(() => import('./SelectCsvModal'));
 
+let lastSearch = '';
+
 const ModelDataCard = ({ model }: { model: ModelSchema }) => {
   const { Loading, setIsLoading } = useLoading();
-
+  const [searchText, setSearchText] = useState('');
   const {
     data: modelDataList,
     isLoading,
@@ -54,7 +57,8 @@ const ModelDataCard = ({ model }: { model: ModelSchema }) => {
     api: getModelDataList,
     pageSize: 8,
     params: {
-      modelId: model._id
+      modelId: model._id,
+      searchText
     }
   });
 
@@ -152,15 +156,39 @@ const ModelDataCard = ({ model }: { model: ModelSchema }) => {
             >
               手动输入
             </MenuItem>
-            <MenuItem onClick={onOpenSelectFileModal}>文本内容 QA 拆分</MenuItem>
+            <MenuItem onClick={onOpenSelectFileModal}>文本/文件 QA 拆分</MenuItem>
             <MenuItem onClick={onOpenSelectUrlModal}>网站内容 QA 拆分</MenuItem>
             <MenuItem onClick={onOpenSelectCsvModal}>csv 问答对导入</MenuItem>
           </MenuList>
         </Menu>
       </Flex>
-      {!!(splitDataLen && splitDataLen > 0) && (
-        <Box fontSize={'xs'}>{splitDataLen}条数据正在拆分...</Box>
-      )}
+      <Flex mt={4}>
+        {/* 拆分数据提示 */}
+        {!!(splitDataLen && splitDataLen > 0) && (
+          <Box fontSize={'xs'}>{splitDataLen}条数据正在拆分...</Box>
+        )}
+        <Box flex={1}></Box>
+        <Input
+          maxW={'240px'}
+          size={'sm'}
+          value={searchText}
+          placeholder="搜索相关问题和答案，回车确认"
+          onChange={(e) => setSearchText(e.target.value)}
+          onBlur={() => {
+            if (searchText === lastSearch) return;
+            getData(1);
+            lastSearch = searchText;
+          }}
+          onKeyDown={(e) => {
+            if (searchText === lastSearch) return;
+            if (e.key === 'Enter') {
+              getData(1);
+              lastSearch = searchText;
+            }
+          }}
+        />
+      </Flex>
+
       <Box mt={4}>
         <TableContainer minH={'500px'}>
           <Table variant={'simple'}>

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

@@ -12,12 +12,13 @@ import {
   SliderThumb,
   SliderMark,
   Tooltip,
-  Button
+  Button,
+  Select
 } from '@chakra-ui/react';
 import { QuestionOutlineIcon } from '@chakra-ui/icons';
 import type { ModelSchema } from '@/types/mongoSchema';
 import { UseFormReturn } from 'react-hook-form';
-import { modelList } from '@/constants/model';
+import { modelList, ModelVectorSearchModeMap } from '@/constants/model';
 import { formatPrice } from '@/utils/user';
 import { useConfirm } from '@/hooks/useConfirm';
 
@@ -53,20 +54,20 @@ const ModelEditForm = ({
               })}
             ></Input>
           </Flex>
-          <Flex alignItems={'center'} mt={4}>
+          <Flex alignItems={'center'} mt={5}>
             <Box flex={'0 0 80px'} w={0}>
               modelId:
             </Box>
             <Box>{getValues('_id')}</Box>
           </Flex>
         </FormControl>
-        <Flex alignItems={'center'} mt={4}>
+        <Flex alignItems={'center'} mt={5}>
           <Box flex={'0 0 80px'} w={0}>
-            底层模型:
+            模型类型:
           </Box>
-          <Box>{getValues('service.modelName')}</Box>
+          <Box>{modelList.find((item) => item.model === getValues('service.modelName'))?.name}</Box>
         </Flex>
-        <Flex alignItems={'center'} mt={4}>
+        <Flex alignItems={'center'} mt={5}>
           <Box flex={'0 0 80px'} w={0}>
             价格:
           </Box>
@@ -79,7 +80,7 @@ const ModelEditForm = ({
           </Box>
         </Flex>
         <Flex mt={5} alignItems={'center'}>
-          <Box flex={'0 0 80px'}>删除:</Box>
+          <Box flex={'0 0 150px'}>删除模型和数据集</Box>
           <Button
             colorScheme={'gray'}
             variant={'outline'}
@@ -89,15 +90,6 @@ const ModelEditForm = ({
             删除模型
           </Button>
         </Flex>
-        {/* <FormControl mt={4}>
-          <Box mb={1}>介绍:</Box>
-          <Textarea
-            rows={5}
-            maxLength={500}
-            {...register('intro')}
-            placeholder={'模型的介绍，仅做展示，不影响模型的效果'}
-          />
-        </FormControl> */}
       </Card>
       <Card p={4}>
         <Box fontWeight={'bold'}>模型效果</Box>
@@ -143,6 +135,20 @@ const ModelEditForm = ({
             </Slider>
           </Flex>
         </FormControl>
+        {canTrain && (
+          <FormControl mt={4}>
+            <Flex alignItems={'center'}>
+              <Box flex={'0 0 70px'}>搜索模式</Box>
+              <Select {...register('search.mode', { required: '搜索模式不能为空' })}>
+                {Object.entries(ModelVectorSearchModeMap).map(([key, { text }]) => (
+                  <option key={key} value={key}>
+                    {text}
+                  </option>
+                ))}
+              </Select>
+            </Flex>
+          </FormControl>
+        )}
         <Box mt={4}>
           <Box mb={1}>系统提示词</Box>
           <Textarea
@@ -151,8 +157,8 @@ const ModelEditForm = ({
             {...register('systemPrompt')}
             placeholder={
               canTrain
-                ? '训练的模型会根据知识库内容，生成一部分系统提示词，因此在对话时需要消耗更多的 tokens。你仍可以增加一些提示词，让其效果更精确。'
-                : '模型默认的 prompt 词，通过调整该内容，可以生成一个限定范围的模型。\n\n注意，改功能会影响对话的整体朝向！'
+                ? '训练的模型会根据知识库内容，生成一部分系统提示词，因此在对话时需要消耗更多的 tokens。你可以增加提示词，让效果更符合预期。例如: \n1. 请根据知识库内容回答用户问题。\n2. 知识库是电影《铃芽之旅》的内容，根据知识库内容回答。无关问题，拒绝回复！'
+                : '模型默认的 prompt 词，通过调整该内容，可以生成一个限定范围的模型。\n注意，改功能会影响对话的整体朝向！'
             }
           />
         </Box>

src/pages/model/detail/index.tsx

@@ -21,7 +21,6 @@ const ModelDetail = ({ modelId }: { modelId: string }) => {
   const { isPc, media } = useScreen();
   const { setLoading } = useGlobalStore();
 
-  // const SelectFileDom = useRef<HTMLInputElement>(null);
   const [model, setModel] = useState<ModelSchema>(defaultModel);
   const formHooks = useForm<ModelSchema>({
     defaultValues: model
@@ -143,6 +142,7 @@ const ModelDetail = ({ modelId }: { modelId: string }) => {
           systemPrompt: data.systemPrompt,
           intro: data.intro,
           temperature: data.temperature,
+          search: data.search,
           service: data.service,
           security: data.security
         });
@@ -242,11 +242,6 @@ const ModelDetail = ({ modelId }: { modelId: string }) => {
       <Grid mt={5} gridTemplateColumns={media('1fr 1fr', '1fr')} gridGap={5}>
         <ModelEditForm formHooks={formHooks} handleDelModel={handleDelModel} canTrain={canTrain} />
 
-        {/* {canTrain && (
-          <Card p={4}>
-            <Training model={model} />
-          </Card>
-        )} */}
         {canTrain && model._id && (
           <Card
             p={4}
@@ -262,11 +257,6 @@ const ModelDetail = ({ modelId }: { modelId: string }) => {
           </Card>
         )}
       </Grid>
-
-      {/* 文件选择 */}
-      {/* <Box position={'absolute'} w={0} h={0} overflow={'hidden'}>
-        <input ref={SelectFileDom} type="file" accept=".jsonl" onChange={startTraining} />
-      </Box> */}
     </>
   );
 };

src/pages/model/list/components/ModelTable.tsx

@@ -16,6 +16,7 @@ import { formatModelStatus } from '@/constants/model';
 import dayjs from 'dayjs';
 import type { ModelSchema } from '@/types/mongoSchema';
 import { useRouter } from 'next/router';
+import { modelList } from '@/constants/model';
 
 const ModelTable = ({
   models = [],
@@ -31,6 +32,15 @@ const ModelTable = ({
       key: 'name',
       dataIndex: 'name'
     },
+    {
+      title: '模型类型',
+      key: 'service',
+      render: (model: ModelSchema) => (
+        <Box fontWeight={'bold'} whiteSpace={'pre-wrap'} maxW={'200px'}>
+          {modelList.find((item) => item.model === model.service.modelName)?.name}
+        </Box>
+      )
+    },
     {
       title: '最后更新时间',
       key: 'updateTime',
@@ -51,15 +61,7 @@ const ModelTable = ({
         </Tag>
       )
     },
-    {
-      title: 'AI模型',
-      key: 'service',
-      render: (item: ModelSchema) => (
-        <Box wordBreak={'break-all'} whiteSpace={'pre-wrap'} maxW={'200px'}>
-          {item.service.modelName}
-        </Box>
-      )
-    },
+
     {
       title: '操作',
       key: 'control',
@@ -69,7 +71,7 @@ const ModelTable = ({
             对话
           </Button>
           <Button
-            colorScheme={'gray'}
+            variant={'outline'}
             onClick={() => router.push(`/model/detail?modelId=${item._id}`)}
           >
             编辑

src/pages/number/components/BillTable.tsx

@@ -30,7 +30,7 @@ const BillTable = () => {
               <Th>类型</Th>
               <Th>内容长度</Th>
               <Th>Tokens 长度</Th>
-              <Th>消费</Th>
+              <Th>金额</Th>
             </Tr>
           </Thead>
           <Tbody fontSize={'sm'}>

src/pages/number/components/PayRecordTable.tsx

@@ -77,7 +77,7 @@ const PayRecordTable = () => {
                 <Th>订单号</Th>
                 <Th>时间</Th>
                 <Th>金额</Th>
-                <Th>消费</Th>
+                <Th>状态</Th>
                 <Th></Th>
               </Tr>
             </Thead>

src/service/errorCode.ts

@@ -8,7 +8,7 @@ export const openaiError: Record<string, string> = {
 };
 export const openaiError2: Record<string, string> = {
   insufficient_quota: 'API 余额不足',
-  invalid_request_error: '输入参数异常'
+  invalid_request_error: 'openai 接口异常'
 };
 export const proxyError: Record<string, boolean> = {
   ECONNABORTED: true,

src/service/events/pushBill.ts

@@ -21,7 +21,7 @@ export const pushChatBill = async ({
 
   try {
     // 计算 token 数量
-    const tokens = Math.floor(encode(text).length * 0.7);
+    const tokens = Math.floor(encode(text).length * 0.75);
 
     console.log(`chat generate success. text len: ${text.length}. token len: ${tokens}`);
 

src/service/models/model.ts

@@ -1,5 +1,7 @@
 import { Schema, model, models, Model as MongoModel } from 'mongoose';
 import { ModelSchema as ModelType } from '@/types/mongoSchema';
+import { ModelVectorSearchModeMap, ModelVectorSearchModeEnum } from '@/constants/model';
+
 const ModelSchema = new Schema({
   userId: {
     type: Schema.Types.ObjectId,
@@ -43,6 +45,13 @@ const ModelSchema = new Schema({
     max: 10,
     default: 4
   },
+  search: {
+    mode: {
+      type: String,
+      enum: Object.keys(ModelVectorSearchModeMap),
+      default: ModelVectorSearchModeEnum.hightSimilarity
+    }
+  },
   service: {
     company: {
       type: String,

src/service/response.ts

@@ -49,6 +49,6 @@ export const jsonRes = <T = any>(
     code,
     statusText: '',
     message: msg,
-    data: data || null
+    data: data !== undefined ? data : null
   });
 };

src/service/utils/tools.ts

@@ -91,7 +91,7 @@ export const openaiChatFilter = (prompts: ChatItemType[], maxTokens: number) =>
   const formatPrompts = prompts.map((item) => ({
     obj: item.obj,
     value: item.value
-      .replace(/[\u3000\u3001\uff01-\uff5e\u3002]/g, ' ') // 中文标点改空格
+      // .replace(/[\u3000\u3001\uff01-\uff5e\u3002]/g, ' ') // 中文标点改空格
       .replace(/\n+/g, '\n') // 连续空行
       .replace(/[^\S\r\n]+/g, ' ') // 连续空白内容
       .trim()

src/styles/reset.scss

@@ -49,6 +49,11 @@ svg {
   background: #999;
 }
 
+input::placeholder,
+textarea::placeholder {
+  font-size: 0.85em;
+}
+
 @media (max-width: 900px) {
   html {
     font-size: 14px;

src/types/model.d.ts

@@ -5,8 +5,9 @@ export interface ModelUpdateParams {
   systemPrompt: string;
   intro: string;
   temperature: number;
-  service: ModelSchema.service;
-  security: ModelSchema.security;
+  search: ModelSchema['search'];
+  service: ModelSchema['service'];
+  security: ModelSchema['security'];
 }
 
 export interface ModelDataItemType {

src/types/mongoSchema.d.ts

@@ -1,5 +1,10 @@
 import type { ChatItemType } from './chat';
-import { ModelStatusEnum, TrainingStatusEnum, ChatModelNameEnum } from '@/constants/model';
+import {
+  ModelStatusEnum,
+  TrainingStatusEnum,
+  ChatModelNameEnum,
+  ModelVectorSearchModeEnum
+} from '@/constants/model';
 import type { DataType } from './data';
 
 export type ServiceName = 'openai';
@@ -32,6 +37,9 @@ export interface ModelSchema {
   updateTime: number;
   trainingTimes: number;
   temperature: number;
+  search: {
+    mode: `${ModelVectorSearchModeEnum}`;
+  };
   service: {
     company: ServiceName;
     trainId: string; // 训练的模型，训练后就是训练的模型id

src/utils/file.ts

@@ -125,7 +125,7 @@ export const fileDownload = ({
   filename: string;
 }) => {
   // 导出为文件
-  const blob = new Blob([text], { type: `${type};charset=utf-8` });
+  const blob = new Blob([`\uFEFF${text}`], { type: `${type};charset=utf-8;` });
 
   // 创建下载链接
   const downloadLink = document.createElement('a');

src/utils/tools.ts

@@ -71,10 +71,3 @@ export const formatVector = (vector: number[]) => {
 
   return formattedVector;
 };
-
-/**
- * 字符串清理，替换换行符号
- */
-export const clearStrLineBreak = (str: string) => {
-  return str.replace(/\n+/g, '\n').replace(/\n/g, '\\n').trim();
-};