From eefc976d4286c9d0569498f1fa09dfe1ce0d4fe2 Mon Sep 17 00:00:00 2001
From: archer <545436317@qq.com>
Date: Sat, 26 Aug 2023 19:21:15 +0800
Subject: [PATCH] perf: qa prompt

---
 README.md                                           | 6 +++---
 client/src/pages/kb/detail/components/Import/QA.tsx | 6 +++---
 client/src/service/events/generateQA.ts             | 8 ++++----
 client/src/utils/file.ts                            | 8 ++++++--
 4 files changed, 16 insertions(+), 12 deletions(-)

diff --git a/README.md b/README.md
index ba08dd567..f4a273d20 100644
--- a/README.md
+++ b/README.md
@@ -85,9 +85,9 @@ FastGPT 是一个基于 LLM 大语言模型的知识库问答系统，提供开
 
 ## 🏘️ 社区交流群
 
-| 交流群                                              | 小助手                                         |
-| --------------------------------------------------- | ---------------------------------------------- |
-| ![](https://otnvvf-imgs.oss.laf.run/wxqun300-2.jpg) | ![](https://otnvvf-imgs.oss.laf.run/wx300.jpg) |
+| 交流群                                            | 小助手                                         |
+| ------------------------------------------------- | ---------------------------------------------- |
+| ![](https://otnvvf-imgs.oss.laf.run/wxqun300.jpg) | ![](https://otnvvf-imgs.oss.laf.run/wx300.jpg) |
 
 ## 👀 其他
 
diff --git a/client/src/pages/kb/detail/components/Import/QA.tsx b/client/src/pages/kb/detail/components/Import/QA.tsx
index d3fc1c609..eb3cc0b7e 100644
--- a/client/src/pages/kb/detail/components/Import/QA.tsx
+++ b/client/src/pages/kb/detail/components/Import/QA.tsx
@@ -187,20 +187,20 @@ const QAImport = ({ kbId }: { kbId: string }) => {
               <Box mb={2}>
                 QA 拆分引导词{' '}
                 <MyTooltip
-                  label={`可输入关于文件内容的范围介绍，例如:\n1. 关于 Laf 的介绍\n2. xxx的简历`}
+                  label={`可输入关于文件内容的范围介绍，例如:\n1. Laf 的介绍\n2. xxx的简历\n最终会补全为: 关于{输入的内容}`}
                   forceShow
                 >
                   <QuestionOutlineIcon ml={1} />
                 </MyTooltip>
               </Box>
               <Flex alignItems={'center'} fontSize={'sm'}>
-                <Box mr={2}>下面是</Box>
+                <Box mr={2}>关于</Box>
                 <Input
                   flex={1}
                   placeholder={'Laf 云函数的介绍'}
                   bg={'myWhite.500'}
                   defaultValue={prompt}
-                  onBlur={(e) => (e.target.value ? setPrompt(`下面是"${e.target.value}"`) : '')}
+                  onBlur={(e) => (e.target.value ? setPrompt(`关于"${e.target.value}"`) : '')}
                 />
               </Flex>
             </Box>
diff --git a/client/src/service/events/generateQA.ts b/client/src/service/events/generateQA.ts
index 2a98b1605..d94547f57 100644
--- a/client/src/service/events/generateQA.ts
+++ b/client/src/service/events/generateQA.ts
@@ -64,14 +64,14 @@ export async function generateQA(): Promise<any> {
         const messages: ChatCompletionRequestMessage[] = [
           {
             role: 'system',
-            content: `你是出题人，${
-              data.prompt || '我会发送一段长文本'
-            }，请从中提取出 25 个问题和答案. 答案详细完整，并按下面格式返回: 
+            content: `我会给你发送一段长文本，${
+              data.prompt ? `是${data.prompt}，` : ''
+            }请学习它，并用 markdown 格式给出 25 个问题和答案，问题可以多样化、自由扩展；答案要详细、解读到位，答案包含普通文本、链接、代码、表格、公示、媒体链接等。按下面 QA 问答格式返回: 
 Q1:
 A1:
 Q2:
 A2:
-...`
+……`
           },
           {
             role: 'user',
diff --git a/client/src/utils/file.ts b/client/src/utils/file.ts
index 32735e280..4afbc49a4 100644
--- a/client/src/utils/file.ts
+++ b/client/src/utils/file.ts
@@ -148,7 +148,7 @@ export const fileDownload = ({
  * maxLen > overlapLen
  */
 export const splitText2Chunks = ({ text, maxLen }: { text: string; maxLen: number }) => {
-  const overlapLen = Math.floor(maxLen * 0.3); // Overlap length
+  const overlapLen = Math.floor(maxLen * 0.25); // Overlap length
 
   try {
     const splitTexts = text.split(/(?<=[。！？；.!?;])/g);
@@ -281,6 +281,10 @@ export const simpleText = (text: string) => {
   text = text.replace(/([\u4e00-\u9fa5])\s+([\u4e00-\u9fa5])/g, '$1$2');
   text = text.replace(/\n{2,}/g, '\n');
   text = text.replace(/\s{2,}/g, ' ');
-  text = text.replace(/[^\x00-\x7F]/g, ' ');
+
+  text = text.replace(/\\x([0-9A-Fa-f]{2})/g, function (match, hex) {
+    return String.fromCharCode(parseInt(hex, 16));
+  });
+
   return text;
 };