feat: use last quote

src/api/chat.ts

@@ -5,7 +5,7 @@ import { RequestPaging } from '../types/index';
 import type { ShareChatSchema } from '@/types/mongoSchema';
 import type { ShareChatEditType } from '@/types/model';
 import { Obj2Query } from '@/utils/tools';
-import { QuoteItemType } from '@/pages/api/openapi/kb/appKbSearch';
+import type { QuoteItemType } from '@/pages/api/openapi/kb/appKbSearch';
 import type { Props as UpdateHistoryProps } from '@/pages/api/chat/history/updateChatHistory';
 
 /**

src/pages/api/chat/chat.ts

@@ -50,6 +50,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
 
     // 读取对话内容
     const prompts = [...content, prompt[0]];
+
     const {
       code = 200,
       systemPrompts = [],
@@ -61,7 +62,8 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
         const { code, searchPrompts, rawSearch, guidePrompt } = await appKbSearch({
           model,
           userId,
-          prompts,
+          fixedQuote: content[content.length - 1]?.quote || [],
+          prompt: prompt[0],
           similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity
         });
 
@@ -114,7 +116,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       return res.end(response);
     }
 
-    prompts.splice(prompts.length - 3, 0, ...systemPrompts);
+    prompts.unshift(...systemPrompts);
 
     // content check
     await sensitiveCheck({

src/pages/api/chat/shareChat/chat.ts

@@ -47,7 +47,8 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
         const { code, searchPrompts } = await appKbSearch({
           model,
           userId,
-          prompts,
+          fixedQuote: [],
+          prompt: prompts[prompts.length - 1],
           similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity
         });
 
@@ -74,7 +75,7 @@ export default async function handler(req: NextApiRequest, res: NextApiResponse)
       return res.send(systemPrompts[0]?.value);
     }
 
-    prompts.splice(prompts.length - 3, 0, ...systemPrompts);
+    prompts.unshift(...systemPrompts);
 
     // content check
     await sensitiveCheck({

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

@@ -75,10 +75,11 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     // 使用了知识库搜索
     if (model.chat.relatedKbs.length > 0) {
       const { code, searchPrompts } = await appKbSearch({
-        prompts,
-        similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity,
         model,
-        userId
+        userId,
+        fixedQuote: [],
+        prompt: prompts[prompts.length - 1],
+        similarity: ModelVectorSearchModeMap[model.chat.searchMode]?.similarity
       });
 
       // search result is empty
@@ -101,7 +102,7 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
       ];
     }
 
-    prompts.splice(prompts.length - 3, 0, ...systemPrompts);
+    prompts.unshift(...systemPrompts);
 
     // content check
     await sensitiveCheck({

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

@@ -49,10 +49,11 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
     });
 
     const result = await appKbSearch({
+      model,
       userId,
-      prompts,
-      similarity,
-      model
+      fixedQuote: [],
+      prompt: prompts[prompts.length - 1],
+      similarity
     });
 
     jsonRes<Response>(res, {
@@ -70,67 +71,53 @@ export default withNextCors(async function handler(req: NextApiRequest, res: Nex
 export async function appKbSearch({
   model,
   userId,
-  prompts,
+  fixedQuote,
+  prompt,
   similarity
 }: {
+  model: ModelSchema;
   userId: string;
-  prompts: ChatItemSimpleType[];
+  fixedQuote: QuoteItemType[];
+  prompt: ChatItemSimpleType;
   similarity: number;
-  model: ModelSchema;
 }): Promise<Response> {
   const modelConstantsData = ChatModelMap[model.chat.chatModel];
 
-  // search two times.
-  const userPrompts = prompts.filter((item) => item.obj === 'Human');
-
-  const input: string[] = [
-    userPrompts[userPrompts.length - 1].value,
-    userPrompts[userPrompts.length - 2]?.value
-  ].filter((item) => item);
-
   // get vector
-  const promptVectors = await openaiEmbedding({
+  const promptVector = await openaiEmbedding({
     userId,
-    input,
+    input: [prompt.value],
     type: 'chat'
   });
 
   // search kb
-  const searchRes = await Promise.all(
-    promptVectors.map((promptVector) =>
-      PgClient.select<QuoteItemType>('modelData', {
-        fields: ['id', 'q', 'a'],
-        where: [
-          `kb_id IN (${model.chat.relatedKbs.map((item) => `'${item}'`).join(',')})`,
-          'AND',
-          `vector <=> '[${promptVector}]' < ${similarity}`
-        ],
-        order: [{ field: 'vector', mode: `<=> '[${promptVector}]'` }],
-        limit: promptVectors.length === 1 ? 15 : 10
-      }).then((res) => res.rows)
-    )
-  );
+  const { rows: searchRes } = await PgClient.select<QuoteItemType>('modelData', {
+    fields: ['id', 'q', 'a'],
+    where: [
+      `kb_id IN (${model.chat.relatedKbs.map((item) => `'${item}'`).join(',')})`,
+      'AND',
+      `vector <=> '[${promptVector[0]}]' < ${similarity}`
+    ],
+    order: [{ field: 'vector', mode: `<=> '[${promptVector[0]}]'` }],
+    limit: 8
+  });
 
   // filter same search result
   const idSet = new Set<string>();
-  const filterSearch = searchRes.map((search) =>
-    search.filter((item) => {
-      if (idSet.has(item.id)) {
-        return false;
-      }
-      idSet.add(item.id);
-      return true;
-    })
-  );
+  const filterSearch = [
+    ...searchRes.slice(0, 3),
+    ...fixedQuote.slice(0, 2),
+    ...searchRes.slice(3),
+    ...fixedQuote.slice(2, 5)
+  ].filter((item) => {
+    if (idSet.has(item.id)) {
+      return false;
+    }
+    idSet.add(item.id);
+    return true;
+  });
 
-  // slice search result by rate.
-  const sliceRateMap: Record<number, number[]> = {
-    1: [1],
-    2: [0.7, 0.3]
-  };
-  const sliceRate = sliceRateMap[searchRes.length] || sliceRateMap[0];
   // 计算固定提示词的 token 数量
-
   const guidePrompt = model.chat.systemPrompt // user system prompt
     ? {
         obj: ChatRoleEnum.System,
@@ -154,24 +141,21 @@ export async function appKbSearch({
   const fixedSystemTokens = modelToolMap[model.chat.chatModel].countTokens({
     messages: [guidePrompt]
   });
-  const maxTokens = modelConstantsData.systemMaxToken - fixedSystemTokens;
-  const sliceResult = sliceRate.map((rate, i) =>
-    modelToolMap[model.chat.chatModel]
-      .tokenSlice({
-        maxToken: Math.round(maxTokens * rate),
-        messages: filterSearch[i].map((item) => ({
-          obj: ChatRoleEnum.System,
-          value: `${item.q}\n${item.a}`
-        }))
-      })
-      .map((item) => item.value)
-  );
+  const sliceResult = modelToolMap[model.chat.chatModel]
+    .tokenSlice({
+      maxToken: modelConstantsData.systemMaxToken - fixedSystemTokens,
+      messages: filterSearch.map((item) => ({
+        obj: ChatRoleEnum.System,
+        value: `${item.q}\n${item.a}`
+      }))
+    })
+    .map((item) => item.value);
 
   // slice filterSearch
-  const sliceSearch = filterSearch.map((item, i) => item.slice(0, sliceResult[i].length)).flat();
+  const rawSearch = filterSearch.slice(0, sliceResult.length);
 
   //  system prompt
-  const systemPrompt = sliceResult.flat().join('\n').trim();
+  const systemPrompt = sliceResult.join('\n').trim();
 
   /* 高相似度+不回复 */
   if (!systemPrompt && model.chat.searchMode === appVectorSearchModeEnum.hightSimilarity) {
@@ -206,7 +190,7 @@ export async function appKbSearch({
 
   return {
     code: 200,
-    rawSearch: sliceSearch,
+    rawSearch,
     guidePrompt: guidePrompt.value || '',
     searchPrompts: [
       {

src/service/utils/auth.ts

@@ -280,7 +280,8 @@ export const authChat = async ({
       {
         $project: {
           obj: '$content.obj',
-          value: '$content.value'
+          value: '$content.value',
+          quote: '$content.quote'
         }
       }
     ]);

src/service/utils/chat/index.ts

@@ -89,39 +89,55 @@ export const ChatContextFilter = ({
   prompts: ChatItemSimpleType[];
   maxTokens: number;
 }) => {
+  const systemPrompts: ChatItemSimpleType[] = [];
+  const chatPrompts: ChatItemSimpleType[] = [];
+
   let rawTextLen = 0;
-  const formatPrompts = prompts.map<ChatItemSimpleType>((item) => {
+  prompts.forEach((item) => {
     const val = simplifyStr(item.value);
     rawTextLen += val.length;
-    return {
+
+    const data = {
       obj: item.obj,
       value: val
     };
+
+    if (item.obj === ChatRoleEnum.System) {
+      systemPrompts.push(data);
+    } else {
+      chatPrompts.push(data);
+    }
   });
 
   // 长度太小时，不需要进行 token 截断
-  if (formatPrompts.length <= 2 || rawTextLen < maxTokens * 0.5) {
-    return formatPrompts;
+  if (rawTextLen < maxTokens * 0.5) {
+    return [...systemPrompts, ...chatPrompts];
   }
 
+  // 去掉 system 的 token
+  maxTokens -= modelToolMap[model].countTokens({
+    messages: systemPrompts
+  });
+
   // 根据 tokens 截断内容
   const chats: ChatItemSimpleType[] = [];
 
   // 从后往前截取对话内容
-  for (let i = formatPrompts.length - 1; i >= 0; i--) {
-    chats.unshift(formatPrompts[i]);
+  for (let i = chatPrompts.length - 1; i >= 0; i--) {
+    chats.unshift(chatPrompts[i]);
 
     const tokens = modelToolMap[model].countTokens({
       messages: chats
     });
 
     /* 整体 tokens 超出范围, system必须保留 */
-    if (tokens >= maxTokens && formatPrompts[i].obj !== ChatRoleEnum.System) {
-      return chats.slice(1);
+    if (tokens >= maxTokens) {
+      chats.shift();
+      break;
     }
   }
 
-  return chats;
+  return [...systemPrompts, ...chats];
 };
 
 /* stream response */