perf: chat上下文截断;QA提示词

3ea2cf1d · archer · 4397a0ad · 3ea2cf1d · 3ea2cf1d · 3ea2cf1d
Commit 3ea2cf1d authored Apr 21, 2023 by archer
10 changed files
--- a/src/constants/model.ts
+++ b/src/constants/model.ts
@@ -35,7 +35,7 @@ export const modelList: ModelConstantsData[] = [
    model: ChatModelNameEnum.GPT35,
    trainName: '',
    maxToken: 4000,
-    contextMaxToken: 7500,
+    contextMaxToken: 7000,
    maxTemperature: 1.5,
    price: 3
  },

--- a/src/pages/api/chat/chatGpt.ts
+++ b/src/pages/api/chat/chatGpt.ts
@@ -61,7 +61,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
    }
    // 控制在 tokens 数量，防止超出
-    // const filterPrompts = openaiChatFilter(prompts, modelConstantsData.contextMaxToken);
+    const filterPrompts = openaiChatFilter(prompts, modelConstantsData.contextMaxToken);
    // 格式化文本内容成 chatgpt 格式
    const map = {
@@ -69,14 +69,25 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      AI: ChatCompletionRequestMessageRoleEnum.Assistant,
      SYSTEM: ChatCompletionRequestMessageRoleEnum.System
    };
-    const formatPrompts: ChatCompletionRequestMessage[] = prompts.map((item: ChatItemType) => ({
+    const formatPrompts: ChatCompletionRequestMessage[] = filterPrompts.map(
-      role: map[item.obj],
+      (item: ChatItemType) => ({
-      content: item.value
+        role: map[item.obj],
-    }));
+        content: item.value
-    // console.log(formatPrompts);
+      })
+    );
    // 计算温度
    const temperature = modelConstantsData.maxTemperature * (model.temperature / 10);
+    // console.log({
+    //   model: model.service.chatModel,
+    //   temperature: temperature,
+    //   // max_tokens: modelConstantsData.maxToken,
+    //   messages: formatPrompts,
+    //   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: true,
+    //   stop: ['.!?。']
+    // });
    // 获取 chatAPI
    const chatAPI = getOpenAIApi(userApiKey || systemKey);
    // 发出请求

--- a/src/pages/api/chat/vectorGpt.ts
+++ b/src/pages/api/chat/vectorGpt.ts
 import type { NextApiRequest, NextApiResponse } from 'next';
 import { connectToDatabase } from '@/service/mongo';
 import { authChat } from '@/service/utils/chat';
-import { httpsAgent, systemPromptFilter } from '@/service/utils/tools';
+import { httpsAgent, systemPromptFilter, openaiChatFilter } from '@/service/utils/tools';
 import { ChatCompletionRequestMessage, ChatCompletionRequestMessageRoleEnum } from 'openai';
 import { ChatItemType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
@@ -79,7 +79,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
        `vector <=> '[${promptVector}]' < ${similarity}`
      ],
      order: [{ field: 'vector', mode: `<=> '[${promptVector}]'` }],
-      limit: 30
+      limit: 20
    });
    const formatRedisPrompt: string[] = vectorSearch.rows.map((item) => `${item.q}\n${item.a}`);
@@ -116,7 +116,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
    }
    // 控制在 tokens 数量，防止超出
-    // const filterPrompts = openaiChatFilter(prompts, modelConstantsData.contextMaxToken);
+    const filterPrompts = openaiChatFilter(prompts, modelConstantsData.contextMaxToken);
    // 格式化文本内容成 chatgpt 格式
    const map = {
@@ -124,10 +124,12 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      AI: ChatCompletionRequestMessageRoleEnum.Assistant,
      SYSTEM: ChatCompletionRequestMessageRoleEnum.System
    };
-    const formatPrompts: ChatCompletionRequestMessage[] = prompts.map((item: ChatItemType) => ({
+    const formatPrompts: ChatCompletionRequestMessage[] = filterPrompts.map(
-      role: map[item.obj],
+      (item: ChatItemType) => ({
-      content: item.value
+        role: map[item.obj],
-    }));
+        content: item.value
+      })
+    );
    // console.log(formatPrompts);
    // 计算温度
    const temperature = modelConstantsData.maxTemperature * (model.temperature / 10);

--- a/src/pages/api/model/data/splitData.ts
+++ b/src/pages/api/model/data/splitData.ts
@@ -41,11 +41,11 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      const tokens = encode(splitText + chunk).length;
      if (tokens >= 4000) {
        // 超过 4000，不要这块内容
-        textList.push(splitText);
+        splitText && textList.push(splitText);
        splitText = chunk;
      } else if (tokens >= 3000) {
        // 超过 3000，取内容
-        textList.push(splitText + chunk);
+        splitText && textList.push(splitText + chunk);
        splitText = '';
      } else {
        //没超过 3000，继续添加

--- a/src/pages/api/openapi/chat/chatGpt.ts
+++ b/src/pages/api/openapi/chat/chatGpt.ts
@@ -74,16 +74,21 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      });
    }
+    // 控制在 tokens 数量，防止超出
+    const filterPrompts = openaiChatFilter(prompts, modelConstantsData.contextMaxToken);
    // 格式化文本内容成 chatgpt 格式
    const map = {
      Human: ChatCompletionRequestMessageRoleEnum.User,
      AI: ChatCompletionRequestMessageRoleEnum.Assistant,
      SYSTEM: ChatCompletionRequestMessageRoleEnum.System
    };
-    const formatPrompts: ChatCompletionRequestMessage[] = prompts.map((item: ChatItemType) => ({
+    const formatPrompts: ChatCompletionRequestMessage[] = filterPrompts.map(
-      role: map[item.obj],
+      (item: ChatItemType) => ({
-      content: item.value
+        role: map[item.obj],
-    }));
+        content: item.value
+      })
+    );
    // console.log(formatPrompts);
    // 计算温度
    const temperature = modelConstantsData.maxTemperature * (model.temperature / 10);

--- a/src/pages/api/openapi/chat/vectorGpt.ts
+++ b/src/pages/api/openapi/chat/vectorGpt.ts
 import type { NextApiRequest, NextApiResponse } from 'next';
 import { connectToDatabase, Model } from '@/service/mongo';
-import { httpsAgent, systemPromptFilter, authOpenApiKey } from '@/service/utils/tools';
+import {
+  httpsAgent,
+  systemPromptFilter,
+  authOpenApiKey,
+  openaiChatFilter
+} from '@/service/utils/tools';
 import { ChatCompletionRequestMessage, ChatCompletionRequestMessageRoleEnum } from 'openai';
 import { ChatItemType } from '@/types/chat';
 import { jsonRes } from '@/service/response';
@@ -93,7 +98,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
        `vector <=> '[${promptVector}]' < ${similarity}`
      ],
      order: [{ field: 'vector', mode: `<=> '[${promptVector}]'` }],
-      limit: 30
+      limit: 20
    });
    const formatRedisPrompt: string[] = vectorSearch.rows.map((item) => `${item.q}\n${item.a}`);
@@ -134,16 +139,21 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
      });
    }
+    // 控制在 tokens 数量，防止超出
+    const filterPrompts = openaiChatFilter(prompts, modelConstantsData.contextMaxToken);
    // 格式化文本内容成 chatgpt 格式
    const map = {
      Human: ChatCompletionRequestMessageRoleEnum.User,
      AI: ChatCompletionRequestMessageRoleEnum.Assistant,
      SYSTEM: ChatCompletionRequestMessageRoleEnum.System
    };
-    const formatPrompts: ChatCompletionRequestMessage[] = prompts.map((item: ChatItemType) => ({
+    const formatPrompts: ChatCompletionRequestMessage[] = filterPrompts.map(
-      role: map[item.obj],
+      (item: ChatItemType) => ({
-      content: item.value
+        role: map[item.obj],
-    }));
+        content: item.value
+      })
+    );
    // console.log(formatPrompts);
    // 计算温度
    const temperature = modelConstantsData.maxTemperature * (model.temperature / 10);

--- a/src/pages/chat/index.tsx
+++ b/src/pages/chat/index.tsx
@@ -88,7 +88,7 @@ const Chat = ({ chatId }: { chatId: string }) => {
    throttle(() => {
      if (!ChatBox.current) return;
      const isBottom =
-        ChatBox.current.scrollTop + ChatBox.current.clientHeight + 80 >=
+        ChatBox.current.scrollTop + ChatBox.current.clientHeight + 150 >=
        ChatBox.current.scrollHeight;
      isBottom && scrollToBottom('auto');

--- a/src/pages/model/detail/components/SelectFileModal.tsx
+++ b/src/pages/model/detail/components/SelectFileModal.tsx
@@ -86,7 +86,7 @@ const SelectFileModal = ({
      await postModelDataSplitData({
        modelId,
        text: fileText.replace(/\\n/g, '\n').replace(/\n+/g, '\n'),
-        prompt: `下面是${prompt || '一段长文本'}`
+        prompt: `下面是"${prompt || '一段长文本'}"`
      });
      toast({
        title: '导入数据成功,需要一段拆解和训练',

--- a/src/pages/model/detail/components/SelectUrlModal.tsx
+++ b/src/pages/model/detail/components/SelectUrlModal.tsx
@@ -45,7 +45,7 @@ const SelectUrlModal = ({
      await postModelDataSplitData({
        modelId,
        text: webText,
-        prompt: `下面是${prompt || '一段长文本'}`
+        prompt: `下面是"${prompt || '一段长文本'}"`
      });
      toast({
        title: '导入数据成功,需要一段拆解和训练',

--- a/src/service/events/generateQA.ts
+++ b/src/service/events/generateQA.ts
@@ -69,9 +69,9 @@ export async function generateQA(next = false): Promise<any> {
    const chatAPI = getOpenAIApi(userApiKey || systemKey);
    const systemPrompt: ChatCompletionRequestMessage = {
      role: 'system',
-      content: `${
+      content: `你是出题官.${
-        dataItem.prompt || '下面是一段长文本'
+        dataItem.prompt || '下面是"一段长文本"'
-      },请从中提取出5至30个问题和答案,并按以下格式返回: Q1:\nA1:\nQ2:\nA2:\n`
+      },从中选出5至20个题目和答案,题目包含问答题,计算题,代码题等.答案要详细.按格式返回: Q1:\nA1:\nQ2:\nA2:\n`
    };
    // 请求 chatgpt 获取回答
@@ -114,7 +114,8 @@ export async function generateQA(next = false): Promise<any> {
            };
          })
          .catch((err) => {
-            console.log('QA 拆分错误', err);
+            console.log('QA拆分错误');
+            console.log(err.response?.status, err.response?.statusText, err.response?.data);
            return Promise.reject(err);
          })
      )