Remove quote prompt in api request (#4724)

* chat completion add parsequote param (#4720)

* chat completion add parsequote param

* fix

* perf: quote prompt

---------

Co-authored-by: heheer <heheer@sealos.io>

packages/global/core/ai/prompt/AIChat.ts

@@ -321,13 +321,24 @@ export const Prompt_systemQuotePromptList: PromptTemplateItem[] = [
   }
 ];
 
-export const getQuotePrompt = (version?: string, role: 'user' | 'system' = 'user') => {
+export const getQuotePrompt = (
+  version?: string,
+  role: 'user' | 'system' = 'user',
+  parseQuote = true
+) => {
   const quotePromptTemplates =
     role === 'user' ? Prompt_userQuotePromptList : Prompt_systemQuotePromptList;
 
   const defaultTemplate = quotePromptTemplates[0].value;
 
-  return getPromptByVersion(version, defaultTemplate);
+  return parseQuote
+    ? getPromptByVersion(version, defaultTemplate)
+    : getPromptByVersion(version, defaultTemplate).replace(
+        `- 使用 [id](QUOTE) 格式来引用<Reference></Reference>中的知识，其中 QUOTE 是固定常量, id 为引文中的 id。
+- 在每段结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](QUOTE)。"
+- 每段至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。`,
+        ''
+      );
 };
 
 // Document quote prompt

packages/global/core/ai/prompt/dataset.ts

@@ -1,5 +1,6 @@
-export const getDatasetSearchToolResponsePrompt = () => {
+export const getDatasetSearchToolResponsePrompt = (parseQuote: boolean) => {
-  return `## Role
+  return parseQuote
+    ? `## Role
 你是一个知识库回答助手，可以 "quotes" 中的内容作为本次对话的参考。为了使回答结果更加可信并且可追溯，你需要在每段话结尾添加引用标记。
 
 ## Rules
@@ -10,5 +11,14 @@ export const getDatasetSearchToolResponsePrompt = () => {
 - 使用与问题相同的语言回答。
 - 使用 [id](QUOTE) 格式来引用 "quotes" 中的知识，其中 QUOTE 是固定常量, id 为引文中的 id。
 - 在每段话结尾自然地整合引用。例如: "FastGPT 是一个基于大语言模型(LLM)的知识库问答系统[67e517e74767063e882d6861](QUOTE)。"
-- 每段话至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。`;
+- 每段话至少包含一个引用，也可根据内容需要加入多个引用，按顺序排列。`
+    : `## Role
+你是一个知识库回答助手，可以 "quotes" 中的内容作为本次对话的参考。
+
+## Rules
+- 如果你不清楚答案，你需要澄清。
+- 避免提及你是从 "quotes" 获取的知识。
+- 保持答案与 "quotes" 中描述的一致。
+- 使用 Markdown 语法优化回答格式。尤其是图片、表格、序列号等内容，需严格完整输出。
+- 使用与问题相同的语言回答。`;
 };

packages/global/core/workflow/runtime/type.d.ts

@@ -58,6 +58,7 @@ export type ChatDispatchProps = {
   chatConfig: AppSchema['chatConfig'];
   lastInteractive?: WorkflowInteractiveResponseType; // last interactive response
   stream: boolean;
+  parseQuote?: boolean;
   maxRunTimes: number;
   isToolCall?: boolean;
   workflowStreamResponse?: WorkflowResponseType;

packages/service/core/workflow/dispatch/chat/oneapi.ts

@@ -75,6 +75,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
     res,
     requestOrigin,
     stream = false,
+    parseQuote = true,
     externalProvider,
     histories,
     node: { name, version },
@@ -158,7 +159,8 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
       userChatInput,
       systemPrompt,
       userFiles,
-      documentQuoteText
+      documentQuoteText,
+      parseQuote
     }),
     // Censor = true and system key, will check content
     (() => {
@@ -450,7 +452,8 @@ async function getChatMessages({
   systemPrompt,
   userChatInput,
   userFiles,
-  documentQuoteText
+  documentQuoteText,
+  parseQuote = true
 }: {
   model: LLMModelItemType;
   maxTokens?: number;
@@ -467,13 +470,16 @@ async function getChatMessages({
 
   userFiles: UserChatItemValueItemType['file'][];
   documentQuoteText?: string; // document quote
+  parseQuote?: boolean;
 }) {
   // Dataset prompt ====>
   // User role or prompt include question
   const quoteRole =
     aiChatQuoteRole === 'user' || datasetQuotePrompt.includes('{{question}}') ? 'user' : 'system';
 
-  const datasetQuotePromptTemplate = datasetQuotePrompt || getQuotePrompt(version, quoteRole);
+  const defaultQuotePrompt = getQuotePrompt(version, quoteRole, parseQuote);
+
+  const datasetQuotePromptTemplate = datasetQuotePrompt || defaultQuotePrompt;
 
   // Reset user input, add dataset quote to user input
   const replaceInputValue =

packages/service/core/workflow/dispatch/dataset/search.ts

@@ -55,6 +55,7 @@ export async function dispatchDatasetSearch(
     runningUserInfo: { tmbId },
     histories,
     node,
+    parseQuote = true,
     params: {
       datasets = [],
       similarity,
@@ -266,7 +267,7 @@ export async function dispatchDatasetSearch(
     [DispatchNodeResponseKeyEnum.nodeResponse]: responseData,
     nodeDispatchUsages,
     [DispatchNodeResponseKeyEnum.toolResponses]: {
-      prompt: getDatasetSearchToolResponsePrompt(),
+      prompt: getDatasetSearchToolResponsePrompt(parseQuote),
       quotes: searchRes.map((item) => ({
         id: item.id,
         sourceName: item.sourceName,

packages/service/core/workflow/dispatch/index.ts

@@ -135,6 +135,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
     timezone,
     externalProvider,
     stream = false,
+    parseQuote = true,
     version = 'v1',
     responseDetail = true,
     responseAllData = true,
@@ -606,6 +607,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
       timezone,
       externalProvider,
       stream,
+      parseQuote,
       node,
       runtimeNodes,
       runtimeEdges,

projects/app/src/pages/api/v1/chat/completions.ts

@@ -74,6 +74,7 @@ export type Props = ChatCompletionCreateParams &
     responseChatItemId?: string;
     stream?: boolean;
     detail?: boolean;
+    parseQuote?: boolean;
     variables: Record<string, any>; // Global variables or plugin inputs
   };
 
@@ -106,6 +107,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
 
     stream = false,
     detail = false,
+    parseQuote = false,
     messages = [],
     variables = {},
     responseChatItemId = getNanoid(),
@@ -289,6 +291,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
             chatConfig,
             histories: newHistories,
             stream,
+            parseQuote,
             maxRunTimes: WORKFLOW_MAX_RUN_TIMES,
             workflowStreamResponse: workflowResponseWrite
           });

projects/app/src/pages/api/v2/chat/completions.ts

@@ -74,6 +74,7 @@ export type Props = ChatCompletionCreateParams &
     responseChatItemId?: string;
     stream?: boolean;
     detail?: boolean;
+    parseQuote?: boolean;
     variables: Record<string, any>; // Global variables or plugin inputs
   };
 
@@ -106,6 +107,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
 
     stream = false,
     detail = false,
+    parseQuote = false,
     messages = [],
     variables = {},
     responseChatItemId = getNanoid(),
@@ -288,6 +290,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
             chatConfig,
             histories: newHistories,
             stream,
+            parseQuote,
             maxRunTimes: WORKFLOW_MAX_RUN_TIMES,
             workflowStreamResponse: workflowResponseWrite,
             version: 'v2',

projects/app/src/web/common/api/fetch.ts

@@ -131,7 +131,8 @@ export const streamFetch = ({
           ...data,
           variables,
           detail: true,
-          stream: true
+          stream: true,
+          parseQuote: true
         })
       };