feat: limit prompt

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/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}`
                 }
               : {})
           }