perf: bill

client/src/pages/api/openapi/kb/appKbSearch.ts

@@ -1,186 +0,0 @@
-import type { NextApiRequest, NextApiResponse } from 'next';
-import { jsonRes } from '@/service/response';
-import { authUser } from '@/service/utils/auth';
-import { PgClient } from '@/service/pg';
-import { withNextCors } from '@/service/utils/tools';
-import type { ChatItemType } from '@/types/chat';
-import type { AppSchema } from '@/types/mongoSchema';
-import { authApp } from '@/service/utils/auth';
-import { ChatModelMap } from '@/constants/model';
-import { ChatRoleEnum } from '@/constants/chat';
-import { openaiEmbedding } from '../plugin/openaiEmbedding';
-import { modelToolMap } from '@/utils/plugin';
-
-export type QuoteItemType = {
-  id: string;
-  q: string;
-  a: string;
-  source?: string;
-};
-type Props = {
-  prompts: ChatItemType[];
-  similarity: number;
-  limit: number;
-  appId: string;
-};
-type Response = {
-  rawSearch: QuoteItemType[];
-  userSystemPrompt: {
-    obj: ChatRoleEnum;
-    value: string;
-  }[];
-  userLimitPrompt: {
-    obj: ChatRoleEnum;
-    value: string;
-  }[];
-  quotePrompt: {
-    obj: ChatRoleEnum;
-    value: string;
-  };
-};
-
-export default withNextCors(async function handler(req: NextApiRequest, res: NextApiResponse<any>) {
-  try {
-    const { userId } = await authUser({ req });
-
-    if (!userId) {
-      throw new Error('userId is empty');
-    }
-
-    const { prompts, similarity, limit, appId } = req.body as Props;
-
-    if (!similarity || !Array.isArray(prompts) || !appId) {
-      throw new Error('params is error');
-    }
-
-    // auth app
-    const { app } = await authApp({
-      appId,
-      userId
-    });
-
-    const result = await appKbSearch({
-      app,
-      userId,
-      fixedQuote: [],
-      prompt: prompts[prompts.length - 1],
-      similarity,
-      limit
-    });
-
-    jsonRes<Response>(res, {
-      data: result
-    });
-  } catch (err) {
-    console.log(err);
-    jsonRes(res, {
-      code: 500,
-      error: err
-    });
-  }
-});
-
-export async function appKbSearch({
-  app,
-  userId,
-  fixedQuote = [],
-  prompt,
-  similarity = 0.8,
-  limit = 5
-}: {
-  app: AppSchema;
-  userId: string;
-  fixedQuote?: QuoteItemType[];
-  prompt: ChatItemType;
-  similarity: number;
-  limit: number;
-}): Promise<Response> {
-  const modelConstantsData = ChatModelMap[app.chat.chatModel];
-
-  // get vector
-  const promptVector = await openaiEmbedding({
-    userId,
-    input: [prompt.value]
-  });
-
-  // search kb
-  const res: any = await PgClient.query(
-    `BEGIN;
-    SET LOCAL ivfflat.probes = ${global.systemEnv.pgIvfflatProbe || 10};
-    select id,q,a,source from modelData where kb_id IN (${app.chat.relatedKbs
-      .map((item) => `'${item}'`)
-      .join(',')}) AND vector <#> '[${promptVector[0]}]' < -${similarity} order by vector <#> '[${
-      promptVector[0]
-    }]' limit ${limit};
-    COMMIT;`
-  );
-
-  const searchRes: QuoteItemType[] = res?.[2]?.rows || [];
-
-  // filter same search result
-  const idSet = new Set<string>();
-  const filterSearch = [
-    ...searchRes.slice(0, 3),
-    ...fixedQuote.slice(0, 2),
-    ...searchRes.slice(3),
-    ...fixedQuote.slice(2, Math.floor(fixedQuote.length * 0.4))
-  ].filter((item) => {
-    if (idSet.has(item.id)) {
-      return false;
-    }
-    idSet.add(item.id);
-    return true;
-  });
-
-  // 计算固定提示词的 token 数量
-  const userSystemPrompt = app.chat.systemPrompt // user system prompt
-    ? [
-        {
-          obj: ChatRoleEnum.System,
-          value: app.chat.systemPrompt
-        }
-      ]
-    : [];
-  const userLimitPrompt = [
-    {
-      obj: ChatRoleEnum.Human,
-      value: app.chat.limitPrompt
-        ? app.chat.limitPrompt
-        : `知识库是关于 ${app.name} 的内容，参考知识库回答问题。与 "${app.name}" 无关内容，直接回复: "我不知道"。`
-    }
-  ];
-
-  const fixedSystemTokens = modelToolMap.countTokens({
-    model: app.chat.chatModel,
-    messages: [...userSystemPrompt, ...userLimitPrompt]
-  });
-
-  // filter part quote by maxToken
-  const sliceResult = modelToolMap
-    .tokenSlice({
-      model: app.chat.chatModel,
-      maxToken: modelConstantsData.systemMaxToken - fixedSystemTokens,
-      messages: filterSearch.map((item, i) => ({
-        obj: ChatRoleEnum.System,
-        value: `${i + 1}: [${item.q}\n${item.a}]`
-      }))
-    })
-    .map((item) => item.value)
-    .join('\n')
-    .trim();
-
-  // slice filterSearch
-  const rawSearch = filterSearch.slice(0, sliceResult.length);
-
-  const quoteText = sliceResult ? `知识库:\n${sliceResult}` : '';
-
-  return {
-    rawSearch,
-    userSystemPrompt,
-    userLimitPrompt,
-    quotePrompt: {
-      obj: ChatRoleEnum.System,
-      value: quoteText
-    }
-  };
-}

client/src/pages/api/openapi/kb/pushData.ts

@@ -15,6 +15,7 @@ type DateItemType = { a: string; q: string; source?: string };
 export type Props = {
   kbId: string;
   data: DateItemType[];
+  model: string;
   mode: `${TrainingModeEnum}`;
   prompt?: string;
 };
@@ -25,14 +26,14 @@ export type Response = {
 
 const modeMaxToken = {
   [TrainingModeEnum.index]: 6000,
-  [TrainingModeEnum.qa]: 10000
+  [TrainingModeEnum.qa]: 12000
 };
 
 export default withNextCors(async function handler(req: NextApiRequest, res: NextApiResponse<any>) {
   try {
-    const { kbId, data, mode, prompt } = req.body as Props;
+    const { kbId, data, mode, prompt, model } = req.body as Props;
 
-    if (!kbId || !Array.isArray(data)) {
+    if (!kbId || !Array.isArray(data) || !model) {
       throw new Error('缺少参数');
     }
     await connectToDatabase();
@@ -46,7 +47,8 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
         data,
         userId,
         mode,
-        prompt
+        prompt,
+        model
       })
     });
   } catch (err) {
@@ -62,7 +64,8 @@ export async function pushDataToKb({
   kbId,
   data,
   mode,
-  prompt
+  prompt,
+  model
 }: { userId: string } & Props): Promise<Response> {
   await authKb({
     userId,
@@ -79,7 +82,7 @@ export async function pushDataToKb({
     if (mode === TrainingModeEnum.qa) {
       // count token
       const token = modelToolMap.countTokens({
-        model: OpenAiChatEnum.GPT3516k,
+        model: 'gpt-3.5-turbo-16k',
         messages: [{ obj: 'System', value: item.q }]
       });
       if (token > modeMaxToken[TrainingModeEnum.qa]) {
@@ -144,6 +147,7 @@ export async function pushDataToKb({
     insertData.map((item) => ({
       q: item.q,
       a: item.a,
+      model,
       source: item.source,
       userId,
       kbId,

client/src/pages/api/openapi/kb/searchTest.ts

@@ -3,7 +3,7 @@ import { jsonRes } from '@/service/response';
 import { authUser } from '@/service/utils/auth';
 import { PgClient } from '@/service/pg';
 import { withNextCors } from '@/service/utils/tools';
-import { openaiEmbedding } from '../plugin/openaiEmbedding';
+import { getVector } from '../plugin/vector';
 import type { KbTestItemType } from '@/types/plugin';
 
 export type Props = {
@@ -27,7 +27,7 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
       throw new Error('缺少用户ID');
     }
 
-    const vector = await openaiEmbedding({
+    const vector = await getVector({
       userId,
       input: [text]
     });

client/src/pages/api/openapi/kb/updateData.ts

@@ -3,7 +3,7 @@ import { jsonRes } from '@/service/response';
 import { authUser } from '@/service/utils/auth';
 import { PgClient } from '@/service/pg';
 import { withNextCors } from '@/service/utils/tools';
-import { openaiEmbedding } from '../plugin/openaiEmbedding';
+import { getVector } from '../plugin/vector';
 
 export default withNextCors(async function handler(req: NextApiRequest, res: NextApiResponse<any>) {
   try {
@@ -19,7 +19,7 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     // get vector
     const vector = await (async () => {
       if (q) {
-        return openaiEmbedding({
+        return getVector({
           userId,
           input: [q]
         });