perf: message filter

src/pages/api/chat/chat.ts

@@ -72,7 +72,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
     } else {
       // 没有用知识库搜索，仅用系统提示词
       model.chat.systemPrompt &&
-        prompts.unshift({
+        prompts.splice(prompts.length - 1, 0, {
           obj: ChatRoleEnum.System,
           value: model.chat.systemPrompt
         });

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

@@ -87,13 +87,12 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       prompts.splice(prompts.length - 1, 0, ...searchPrompts);
     } else {
       // 没有用知识库搜索，仅用系统提示词
-      if (model.chat.systemPrompt) {
+      model.chat.systemPrompt &&
-        prompts.unshift({
+        prompts.splice(prompts.length - 1, 0, {
           obj: ChatRoleEnum.System,
           value: model.chat.systemPrompt
         });
     }
-    }
 
     // 计算温度
     const temperature = (modelConstantsData.maxTemperature * (model.chat.temperature / 10)).toFixed(

src/service/plugins/searchKb.ts

@@ -106,7 +106,7 @@ export const searchKb = async ({
           {
             obj: ChatRoleEnum.System,
             value: `我们来玩问答游戏,规则为:
-1.你忘记你已有的知识
+1.你完全忘记你已有的知识
 2.你只能回答关于"${model.name}"的问题
 3.你只能从知识库中选择内容进行回答
 4.如果问题不在知识库中,你会回答"我不知道。"

src/service/utils/chat/index.ts

@@ -109,35 +109,22 @@ export const ChatContextFilter = ({
 
   // 根据 tokens 截断内容
   const chats: ChatItemSimpleType[] = [];
-  let systemPrompt: ChatItemSimpleType | null = null;
-
-  //  System 词保留
-  if (formatPrompts[0].obj === ChatRoleEnum.System) {
-    const prompt = formatPrompts.shift();
-    if (prompt) {
-      systemPrompt = prompt;
-    }
-  }
-
-  let messages: ChatItemSimpleType[] = [];
 
   // 从后往前截取对话内容
   for (let i = formatPrompts.length - 1; i >= 0; i--) {
     chats.unshift(formatPrompts[i]);
 
-    messages = systemPrompt ? [systemPrompt, ...chats] : chats;
-
     const tokens = modelToolMap[model].countTokens({
-      messages
+      messages: chats
     });
 
     /* 整体 tokens 超出范围 */
     if (tokens >= maxTokens) {
-      return systemPrompt ? [systemPrompt, ...chats.slice(1)] : chats.slice(1);
+      return chats.slice(1);
     }
   }
 
-  return messages;
+  return chats;
 };
 
 /* stream response */