fix: v1 api

README.md

@@ -44,9 +44,10 @@ Fast GPT 允许你使用自己的 openai API KEY 来快速的调用 openai 接
 
 ## Powered by
 
-- [TuShan 5 分钟搭建后台管理系统](https://github.com/msgbyte/tushan)
-- [Laf 3 分钟快速接入三方应用](https://github.com/labring/laf)
-- [Sealos 快速部署集群应用](https://github.com/labring/sealos)
+- [TuShan: 5 分钟搭建后台管理系统](https://github.com/msgbyte/tushan)
+- [Laf: 3 分钟快速接入三方应用](https://github.com/labring/laf)
+- [Sealos: 快速部署集群应用](https://github.com/labring/sealos)
+- [One API: 令牌管理 & 二次分发，支持 Azure](https://github.com/songquanpeng/one-api)
 
 ## 🌟 Star History
 

client/src/api/fetch.ts

@@ -9,90 +9,95 @@ interface StreamFetchProps {
   abortSignal: AbortController;
 }
 export const streamFetch = ({ data, onMessage, abortSignal }: StreamFetchProps) =>
-  new Promise<ChatResponseType & { responseText: string }>(async (resolve, reject) => {
-    try {
-      const response = await window.fetch('/api/openapi/v1/chat/completions', {
-        method: 'POST',
-        headers: {
-          'Content-Type': 'application/json'
-        },
-        signal: abortSignal.signal,
-        body: JSON.stringify({
-          ...data,
-          stream: true
-        })
-      });
+  new Promise<ChatResponseType & { responseText: string; errMsg: string }>(
+    async (resolve, reject) => {
+      try {
+        const response = await window.fetch('/api/openapi/v1/chat/completions', {
+          method: 'POST',
+          headers: {
+            'Content-Type': 'application/json'
+          },
+          signal: abortSignal.signal,
+          body: JSON.stringify({
+            ...data,
+            stream: true
+          })
+        });
 
-      if (response.status !== 200) {
-        const err = await response.json();
-        return reject(err);
-      }
+        if (response.status !== 200) {
+          const err = await response.json();
+          return reject(err);
+        }
 
-      if (!response?.body) {
-        throw new Error('Request Error');
-      }
+        if (!response?.body) {
+          throw new Error('Request Error');
+        }
 
-      const reader = response.body?.getReader();
+        const reader = response.body?.getReader();
 
-      // response data
-      let responseText = '';
-      let newChatId = '';
-      let quoteLen = 0;
+        // response data
+        let responseText = '';
+        let newChatId = '';
+        let quoteLen = 0;
+        let errMsg = '';
 
-      const read = async () => {
-        try {
-          const { done, value } = await reader.read();
-          if (done) {
-            if (response.status === 200) {
+        const read = async () => {
+          try {
+            const { done, value } = await reader.read();
+            if (done) {
+              if (response.status === 200) {
+                return resolve({
+                  responseText,
+                  newChatId,
+                  quoteLen,
+                  errMsg
+                });
+              } else {
+                return reject('响应过程出现异常~');
+              }
+            }
+            const chunkResponse = parseStreamChunk(value);
+
+            chunkResponse.forEach((item) => {
+              // parse json data
+              const data = (() => {
+                try {
+                  return JSON.parse(item.data);
+                } catch (error) {
+                  return item.data;
+                }
+              })();
+
+              if (item.event === sseResponseEventEnum.answer && data !== '[DONE]') {
+                const answer: string = data?.choices?.[0].delta.content || '';
+                onMessage(answer);
+                responseText += answer;
+              } else if (item.event === sseResponseEventEnum.chatResponse) {
+                const chatResponse = data as ChatResponseType;
+                newChatId = chatResponse.newChatId;
+                quoteLen = chatResponse.quoteLen || 0;
+              } else if (item.event === sseResponseEventEnum.error) {
+                errMsg = getErrText(data, '流响应错误');
+              }
+            });
+            read();
+          } catch (err: any) {
+            if (err?.message === 'The user aborted a request.') {
               return resolve({
                 responseText,
                 newChatId,
-                quoteLen
+                quoteLen,
+                errMsg
               });
-            } else {
-              return reject('响应过程出现异常~');
             }
+            reject(getErrText(err, '请求异常'));
           }
-          const chunkResponse = parseStreamChunk(value);
+        };
+        read();
+      } catch (err: any) {
+        console.log(err);
 
-          chunkResponse.forEach((item) => {
-            // parse json data
-            const data = (() => {
-              try {
-                return JSON.parse(item.data);
-              } catch (error) {
-                return item.data;
-              }
-            })();
-
-            if (item.event === sseResponseEventEnum.answer && data !== '[DONE]') {
-              const answer: string = data?.choices?.[0].delta.content || '';
-              onMessage(answer);
-              responseText += answer;
-            } else if (item.event === sseResponseEventEnum.chatResponse) {
-              const chatResponse = data as ChatResponseType;
-              newChatId = chatResponse.newChatId;
-              quoteLen = chatResponse.quoteLen || 0;
-            } else if (item.event === sseResponseEventEnum.error) {
-              return reject(getErrText(data, '流响应错误'));
-            }
-          });
-          read();
-        } catch (err: any) {
-          if (err?.message === 'The user aborted a request.') {
-            return resolve({
-              responseText,
-              newChatId,
-              quoteLen
-            });
-          }
-          reject(getErrText(err, '请求异常'));
-        }
-      };
-      read();
-    } catch (err: any) {
-      console.log(err);
-
-      reject(getErrText(err, '请求异常'));
+        reject(getErrText(err, '请求异常'));
+      }
     }
-  });
+  );

client/src/api/response/chat.d.ts

@@ -5,6 +5,7 @@ export interface InitChatResponse {
   chatId: string;
   modelId: string;
   systemPrompt?: string;
+  limitPrompt?: string;
   model: {
     name: string;
     avatar: string;

client/src/components/Select/index.tsx

@@ -26,7 +26,7 @@ const MySelect = ({ placeholder, value, width = 'auto', list, onchange, ...props
 
   return (
     <Menu autoSelect={false} onOpen={onOpen} onClose={onClose}>
-      <MenuButton as={'span'}>
+      <MenuButton style={{ width: '100%' }} as={'span'}>
         <Button
           width={width}
           px={3}

client/src/constants/model.ts

@@ -82,6 +82,7 @@ export const defaultModel: ModelSchema = {
     searchLimit: 5,
     searchEmptyText: '',
     systemPrompt: '',
+    limitPrompt: '',
     temperature: 0,
     maxToken: 4000,
     chatModel: OpenAiChatEnum.GPT35

client/src/pages/api/chat/init.ts

@@ -101,6 +101,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
         },
         chatModel: model.chat.chatModel,
         systemPrompt: isOwner ? model.chat.systemPrompt : '',
+        limitPrompt: isOwner ? model.chat.limitPrompt : '',
         history
       }
     });

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

@@ -65,10 +65,14 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     const modelConstantsData = ChatModelMap[model.chat.chatModel];
     const prompt = prompts[prompts.length - 1];
 
-    const { userSystemPrompt = [], quotePrompt = [] } = await (async () => {
+    const {
+      userSystemPrompt = [],
+      userLimitPrompt = [],
+      quotePrompt = []
+    } = await (async () => {
       // 使用了知识库搜索
       if (model.chat.relatedKbs?.length > 0) {
-        const { userSystemPrompt, quotePrompt } = await appKbSearch({
+        const { quotePrompt, userSystemPrompt, userLimitPrompt } = await appKbSearch({
           model,
           userId,
           fixedQuote: [],
@@ -78,21 +82,29 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
         });
 
         return {
-          userSystemPrompt: userSystemPrompt ? [userSystemPrompt] : [],
+          userSystemPrompt,
+          userLimitPrompt,
           quotePrompt: [quotePrompt]
         };
       }
-      if (model.chat.systemPrompt) {
-        return {
-          userSystemPrompt: [
-            {
-              obj: ChatRoleEnum.System,
-              value: model.chat.systemPrompt
-            }
-          ]
-        };
-      }
-      return {};
+      return {
+        userSystemPrompt: model.chat.systemPrompt
+          ? [
+              {
+                obj: ChatRoleEnum.System,
+                value: model.chat.systemPrompt
+              }
+            ]
+          : [],
+        userLimitPrompt: model.chat.limitPrompt
+          ? [
+              {
+                obj: ChatRoleEnum.Human,
+                value: model.chat.limitPrompt
+              }
+            ]
+          : []
+      };
     })();
 
     // search result is empty
@@ -102,7 +114,13 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     }
 
     // 读取对话内容
-    const completePrompts = [...quotePrompt, ...prompts.slice(0, -1), ...userSystemPrompt, prompt];
+    const completePrompts = [
+      ...quotePrompt,
+      ...userSystemPrompt,
+      ...prompts.slice(0, -1),
+      ...userLimitPrompt,
+      prompt
+    ];
 
     // 计算温度
     const temperature = (modelConstantsData.maxTemperature * (model.chat.temperature / 10)).toFixed(

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

@@ -28,7 +28,11 @@ type Response = {
   userSystemPrompt: {
     obj: ChatRoleEnum;
     value: string;
-  };
+  }[];
+  userLimitPrompt: {
+    obj: ChatRoleEnum;
+    value: string;
+  }[];
   quotePrompt: {
     obj: ChatRoleEnum;
     value: string;
@@ -130,17 +134,24 @@ export async function appKbSearch({
 
   // 计算固定提示词的 token 数量
   const userSystemPrompt = model.chat.systemPrompt // user system prompt
-    ? {
-        obj: ChatRoleEnum.Human,
-        value: model.chat.systemPrompt
-      }
-    : {
-        obj: ChatRoleEnum.Human,
-        value: `知识库是关于 ${model.name} 的内容，参考知识库回答问题。与 "${model.name}" 无关内容，直接回复: "我不知道"。`
-      };
+    ? [
+        {
+          obj: ChatRoleEnum.System,
+          value: model.chat.systemPrompt
+        }
+      ]
+    : [];
+  const userLimitPrompt = [
+    {
+      obj: ChatRoleEnum.Human,
+      value: model.chat.limitPrompt
+        ? model.chat.limitPrompt
+        : `知识库是关于 ${model.name} 的内容，参考知识库回答问题。与 "${model.name}" 无关内容，直接回复: "我不知道"。`
+    }
+  ];
 
   const fixedSystemTokens = modelToolMap[model.chat.chatModel].countTokens({
-    messages: [userSystemPrompt]
+    messages: [...userSystemPrompt, ...userLimitPrompt]
   });
 
   // filter part quote by maxToken
@@ -164,6 +175,7 @@ export async function appKbSearch({
   return {
     rawSearch,
     userSystemPrompt,
+    userLimitPrompt,
     quotePrompt: {
       obj: ChatRoleEnum.System,
       value: quoteText

client/src/pages/api/openapi/v1/chat/completions.ts

@@ -108,11 +108,12 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     const {
       rawSearch = [],
       userSystemPrompt = [],
+      userLimitPrompt = [],
       quotePrompt = []
     } = await (async () => {
       // 使用了知识库搜索
       if (model.chat.relatedKbs?.length > 0) {
-        const { rawSearch, userSystemPrompt, quotePrompt } = await appKbSearch({
+        const { rawSearch, quotePrompt, userSystemPrompt, userLimitPrompt } = await appKbSearch({
           model,
           userId,
           fixedQuote: history[history.length - 1]?.quote,
@@ -123,21 +124,29 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
 
         return {
           rawSearch,
-          userSystemPrompt: userSystemPrompt ? [userSystemPrompt] : [],
+          userSystemPrompt,
+          userLimitPrompt,
           quotePrompt: [quotePrompt]
         };
       }
-      if (model.chat.systemPrompt) {
-        return {
-          userSystemPrompt: [
-            {
-              obj: ChatRoleEnum.System,
-              value: model.chat.systemPrompt
-            }
-          ]
-        };
-      }
-      return {};
+      return {
+        userSystemPrompt: model.chat.systemPrompt
+          ? [
+              {
+                obj: ChatRoleEnum.System,
+                value: model.chat.systemPrompt
+              }
+            ]
+          : [],
+        userLimitPrompt: model.chat.limitPrompt
+          ? [
+              {
+                obj: ChatRoleEnum.Human,
+                value: model.chat.limitPrompt
+              }
+            ]
+          : []
+      };
     })();
 
     // search result is empty
@@ -167,7 +176,13 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     }
 
     // api messages. [quote,context,systemPrompt,question]
-    const completePrompts = [...quotePrompt, ...prompts.slice(0, -1), ...userSystemPrompt, prompt];
+    const completePrompts = [
+      ...quotePrompt,
+      ...userSystemPrompt,
+      ...prompts.slice(0, -1),
+      ...userLimitPrompt,
+      prompt
+    ];
     // chat temperature
     const modelConstantsData = ChatModelMap[model.chat.chatModel];
     // FastGpt temperature range: 1~10
@@ -176,7 +191,7 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     );
 
     await sensitiveCheck({
-      input: `${prompt.value}`
+      input: `${userSystemPrompt[0]?.value}\n${userLimitPrompt[0]?.value}\n${prompt.value}`
     });
 
     // start model api. responseText and totalTokens: valid only if stream = false
@@ -259,7 +274,7 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
             ...(showAppDetail
               ? {
                   quote: rawSearch,
-                  systemPrompt: userSystemPrompt?.[0]?.value
+                  systemPrompt: `${userSystemPrompt[0]?.value}\n\n${userLimitPrompt[0]?.value}`
                 }
               : {})
           }

client/src/pages/chat/index.tsx

@@ -174,7 +174,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
       const messages = adaptChatItem_openAI({ messages: prompts, reserveId: true });
 
       // 流请求，获取数据
-      const { newChatId, quoteLen } = await streamFetch({
+      const { newChatId, quoteLen, errMsg } = await streamFetch({
         data: {
           messages,
           chatId,
@@ -219,7 +219,9 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
             ...item,
             status: 'finish',
             quoteLen,
-            systemPrompt: chatData.systemPrompt
+            systemPrompt: `${chatData.systemPrompt}${`${
+              chatData.limitPrompt ? `\n\n${chatData.limitPrompt}` : ''
+            }`}`
           };
         })
       }));
@@ -230,17 +232,26 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
         loadHistory({ pageNum: 1, init: true });
         loadMyModels(true);
       }, 100);
+
+      if (errMsg) {
+        toast({
+          status: 'warning',
+          title: errMsg
+        });
+      }
     },
     [
       chatId,
       modelId,
-      chatData.systemPrompt,
       setChatData,
-      loadHistory,
-      loadMyModels,
       generatingMessage,
       setForbidLoadChatData,
-      router
+      router,
+      chatData.systemPrompt,
+      chatData.limitPrompt,
+      loadHistory,
+      loadMyModels,
+      toast
     ]
   );
 
@@ -749,7 +760,7 @@ const Chat = ({ modelId, chatId }: { modelId: string; chatId: string }) => {
                                 px={[2, 4]}
                                 onClick={() => setShowSystemPrompt(item.systemPrompt || '')}
                               >
-                                提示词
+                                提示词 & 限定词
                               </Button>
                             )}
                             {!!item.quoteLen && (

client/src/pages/model/components/detail/components/Settings.tsx

@@ -1,5 +1,15 @@
 import React, { useCallback, useState, useMemo } from 'react';
-import { Box, Flex, Button, FormControl, Input, Textarea, Divider } from '@chakra-ui/react';
+import {
+  Box,
+  Flex,
+  Button,
+  FormControl,
+  Input,
+  Textarea,
+  Divider,
+  Tooltip
+} from '@chakra-ui/react';
+import { QuestionOutlineIcon } from '@chakra-ui/icons';
 import { useQuery } from '@tanstack/react-query';
 import { useForm } from 'react-hook-form';
 import { useRouter } from 'next/router';
@@ -20,6 +30,11 @@ import Avatar from '@/components/Avatar';
 import MySelect from '@/components/Select';
 import MySlider from '@/components/Slider';
 
+const systemPromptTip =
+  '模型固定的引导词，通过调整该内容，可以引导模型聊天方向。该内容会被固定在上下文的开头。';
+const limitPromptTip =
+  '限定模型对话范围，会被放置在本次提问前，拥有强引导和限定性。例如:\n1. 知识库是关于 Laf 的介绍，参考知识库回答问题，与 "Laf" 无关内容，直接回复: "我不知道"。\n2. 你仅回答关于 "xxx" 的问题，其他问题回复: "xxxx"';
+
 const Settings = ({ modelId }: { modelId: string }) => {
   const { toast } = useToast();
   const router = useRouter();
@@ -60,7 +75,7 @@ const Settings = ({ modelId }: { modelId: string }) => {
     }
 
     return max;
-  }, [getValues, setValue, refresh]);
+  }, [getValues, setValue]);
 
   // 提交保存模型修改
   const saveSubmitSuccess = useCallback(
@@ -211,7 +226,7 @@ const Settings = ({ modelId }: { modelId: string }) => {
           介绍
         </Box>
         <Textarea
-          rows={5}
+          rows={4}
           maxLength={500}
           placeholder={'给你的 AI 应用一个介绍'}
           {...register('intro')}
@@ -225,11 +240,14 @@ const Settings = ({ modelId }: { modelId: string }) => {
           对话模型
         </Box>
         <MySelect
-          width={['200px', '240px']}
+          width={['100%', '280px']}
           value={getValues('chat.chatModel')}
           list={chatModelList.map((item) => ({
             id: item.chatModel,
-            label: item.name
+            label: `${item.name} (${formatPrice(
+              ChatModelMap[getValues('chat.chatModel')]?.price,
+              1000
+            )} 元/1k tokens)`
           }))}
           onchange={(val: any) => {
             setValue('chat.chatModel', val);
@@ -237,15 +255,6 @@ const Settings = ({ modelId }: { modelId: string }) => {
           }}
         />
       </Flex>
-      <Flex alignItems={'center'} mt={5}>
-        <Box w={['60px', '100px', '140px']} flexShrink={0}>
-          价格
-        </Box>
-        <Box fontSize={['sm', 'md']}>
-          {formatPrice(ChatModelMap[getValues('chat.chatModel')]?.price, 1000)}
-          元/1K tokens(包括上下文和回答)
-        </Box>
-      </Flex>
       <Flex alignItems={'center'} my={10}>
         <Box w={['60px', '100px', '140px']} flexShrink={0}>
           温度
@@ -269,7 +278,7 @@ const Settings = ({ modelId }: { modelId: string }) => {
       </Flex>
       <Flex alignItems={'center'} mt={12} mb={10}>
         <Box w={['60px', '100px', '140px']} flexShrink={0}>
-          最大长度
+          回复上限
         </Box>
         <Box flex={1} ml={'10px'}>
           <MySlider
@@ -292,15 +301,29 @@ const Settings = ({ modelId }: { modelId: string }) => {
       <Flex mt={10} alignItems={'flex-start'}>
         <Box w={['60px', '100px', '140px']} flexShrink={0}>
           提示词
+          <Tooltip label={systemPromptTip}>
+            <QuestionOutlineIcon display={['none', 'inline']} ml={1} />
+          </Tooltip>
         </Box>
         <Textarea
           rows={8}
-          placeholder={
-            '模型默认的 prompt 词，通过调整该内容，可以引导模型聊天方向。\n\n如果使用了知识库搜索，没有填写该内容时，系统会自动补充提示词；如果填写了内容，则以填写的内容为准。'
-          }
+          placeholder={systemPromptTip}
           {...register('chat.systemPrompt')}
         ></Textarea>
       </Flex>
+      <Flex mt={5} alignItems={'flex-start'}>
+        <Box w={['60px', '100px', '140px']} flexShrink={0}>
+          限定词
+          <Tooltip label={limitPromptTip}>
+            <QuestionOutlineIcon display={['none', 'inline']} ml={1} />
+          </Tooltip>
+        </Box>
+        <Textarea
+          rows={5}
+          placeholder={limitPromptTip}
+          {...register('chat.limitPrompt')}
+        ></Textarea>
+      </Flex>
 
       <Flex mt={5} alignItems={'center'}>
         <Box w={['60px', '100px', '140px']} flexShrink={0}></Box>

client/src/service/models/model.ts

@@ -43,7 +43,10 @@ const ModelSchema = new Schema({
       default: ''
     },
     systemPrompt: {
-      // 系统提示词
+      type: String,
+      default: ''
+    },
+    limitPrompt: {
       type: String,
       default: ''
     },

client/src/service/utils/chat/index.ts

@@ -238,6 +238,7 @@ export const V2_StreamResponse = async ({
   }
 
   if (error) {
+    console.log(error);
     return Promise.reject(error);
   }
 

client/src/service/utils/chat/openai.ts

@@ -49,8 +49,8 @@ export const chatResponse = async ({
       messages: adaptMessages,
       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,
-      stop: ['.!?。']
+      stream
+      // stop: ['.!?。']
     },
     {
       timeout: stream ? 60000 : 480000,

client/src/types/mongoSchema.d.ts

@@ -43,6 +43,7 @@ export interface ModelSchema {
     searchLimit: number;
     searchEmptyText: string;
     systemPrompt: string;
+    limitPrompt: string;
     temperature: number;
     maxToken: number;
     chatModel: ChatModelType; // 聊天时用的模型，训练后就是训练的模型

client/src/utils/tools.ts

@@ -116,7 +116,7 @@ export const voiceBroadcast = ({ text }: { text: string }) => {
 };
 
 export const getErrText = (err: any, def = '') => {
-  const msg = typeof err === 'string' ? err : err?.message || def || '';
+  const msg: string = typeof err === 'string' ? err : err?.message || def || '';
   msg && console.log('error =>', msg);
   return msg;
 };