perf: token split text

src/service/events/generateQA.ts

@@ -106,7 +106,7 @@ A2:
           )
           .then((res) => {
             const rawContent = res?.data.choices[0].message?.content || ''; // chatgpt 原本的回复
-            const result = splitText(res?.data.choices[0].message?.content || ''); // 格式化后的QA对
+            const result = formatSplitText(res?.data.choices[0].message?.content || ''); // 格式化后的QA对
             console.log(`split result length: `, result.length);
             // 计费
             pushSplitDataBill({
@@ -190,7 +190,7 @@ A2:
 /**
  * 检查文本是否按格式返回
  */
-function splitText(text: string) {
+function formatSplitText(text: string) {
   const regex = /Q\d+:(\s*)(.*)(\s*)A\d+:(\s*)([\s\S]*?)(?=Q|$)/g; // 匹配Q和A的正则表达式
   const matches = text.matchAll(regex); // 获取所有匹配到的结果
 

src/service/utils/tools.ts

@@ -1,7 +1,7 @@
 import crypto from 'crypto';
 import jwt from 'jsonwebtoken';
 import { ChatItemSimpleType } from '@/types/chat';
-import { countChatTokens } from '@/utils/tools';
+import { countChatTokens, sliceTextByToken } from '@/utils/tools';
 import { ChatCompletionRequestMessageRoleEnum, ChatCompletionRequestMessage } from 'openai';
 import { ChatModelEnum } from '@/constants/model';
 
@@ -111,18 +111,11 @@ export const systemPromptFilter = ({
   prompts: string[];
   maxTokens: number;
 }) => {
-  let splitText = '';
+  const systemPrompt = prompts.join('\n');
 
-  // 从前往前截取
-  for (let i = 0; i < prompts.length; i++) {
-    const prompt = simplifyStr(prompts[i]);
-
-    splitText += `${prompt}\n`;
-    const tokens = countChatTokens({ model, messages: [{ role: 'system', content: splitText }] });
-    if (tokens >= maxTokens) {
-      break;
-    }
-  }
-
-  return splitText.slice(0, splitText.length - 1);
+  return sliceTextByToken({
+    model,
+    text: systemPrompt,
+    length: maxTokens
+  });
 };