feat: dataset quote role support system; fix: adapt o1 model (#2733)

* feat: dataset quote support system role * perf: adapt dataset quote role * fix: adapt o1 model
2025-07-22 20:37:48 +00:00 · 2024-09-18 13:38:50 +08:00
parent 539bc77934
commit 093bfa2134
35 changed files with 582 additions and 268 deletions
--- a/packages/service/core/ai/functions/createQuestionGuide.ts
+++ b/packages/service/core/ai/functions/createQuestionGuide.ts
@@ -2,6 +2,7 @@ import type { ChatCompletionMessageParam } from '@fastgpt/global/core/ai/type.d'
 import { getAIApi } from '../config';
 import { countGptMessagesTokens } from '../../../common/string/tiktoken/index';
 import { loadRequestMessages } from '../../chat/utils';
+import { llmCompletionsBodyFormat } from '../utils';

 export const Prompt_QuestionGuide = `你是一个AI智能助手，可以回答和解决我的问题。请结合前面的对话记录，帮我生成 3 个问题，引导我继续提问，生成问题的语言要与原问题相同。问题的长度应小于20个字符，按 JSON 格式返回: ["问题1", "问题2", "问题3"]`;

@@ -23,16 +24,21 @@ export async function createQuestionGuide({
  const ai = getAIApi({
    timeout: 480000
  });
-  const data = await ai.chat.completions.create({
-    model: model,
-    temperature: 0.1,
-    max_tokens: 200,
-    messages: await loadRequestMessages({
-      messages: concatMessages,
-      useVision: false
-    }),
-    stream: false
-  });
+  const data = await ai.chat.completions.create(
+    llmCompletionsBodyFormat(
+      {
+        model,
+        temperature: 0.1,
+        max_tokens: 200,
+        messages: await loadRequestMessages({
+          messages: concatMessages,
+          useVision: false
+        }),
+        stream: false
+      },
+      model
+    )
+  );

  const answer = data.choices?.[0]?.message?.content || '';

--- a/packages/service/core/ai/functions/queryExtension.ts
+++ b/packages/service/core/ai/functions/queryExtension.ts
@@ -2,9 +2,10 @@ import { replaceVariable } from '@fastgpt/global/common/string/tools';
 import { getAIApi } from '../config';
 import { ChatItemType } from '@fastgpt/global/core/chat/type';
 import { countGptMessagesTokens } from '../../../common/string/tiktoken/index';
-import { ChatCompletionMessageParam } from '@fastgpt/global/core/ai/type';
+import { ChatCompletion, ChatCompletionMessageParam } from '@fastgpt/global/core/ai/type';
 import { chatValue2RuntimePrompt } from '@fastgpt/global/core/chat/adapt';
 import { getLLMModel } from '../model';
+import { llmCompletionsBodyFormat } from '../utils';

 /* 
    query extension - 问题扩展
@@ -150,14 +151,19 @@ A: ${chatBg}
      })
    }
  ] as ChatCompletionMessageParam[];
-  const result = await ai.chat.completions.create({
-    model: modelData.model,
-    temperature: 0.01,
-    // @ts-ignore
-    messages,
-    stream: false,
-    ...modelData.defaultConfig
-  });
+
+  const result = (await ai.chat.completions.create(
+    llmCompletionsBodyFormat(
+      {
+        stream: false,
+        model: modelData.model,
+        temperature: 0.01,
+        // @ts-ignore
+        messages
+      },
+      modelData
+    )
+  )) as ChatCompletion;

  let answer = result.choices?.[0]?.message?.content || '';
  if (!answer) {
--- a/packages/service/core/ai/utils.ts
+++ b/packages/service/core/ai/utils.ts
@@ -1,6 +1,11 @@
 import { LLMModelItemType } from '@fastgpt/global/core/ai/model.d';
-import { ChatCompletionMessageParam } from '@fastgpt/global/core/ai/type';
+import {
+  ChatCompletionCreateParamsNonStreaming,
+  ChatCompletionCreateParamsStreaming,
+  ChatCompletionMessageParam
+} from '@fastgpt/global/core/ai/type';
 import { countGptMessagesTokens } from '../../common/string/tiktoken';
+import { getLLMModel } from './model';

 export const computedMaxToken = async ({
  maxToken,
@@ -32,8 +37,49 @@ export const computedTemperature = ({
  model: LLMModelItemType;
  temperature: number;
 }) => {
+  if (temperature < 1) return temperature;
+
  temperature = +(model.maxTemperature * (temperature / 10)).toFixed(2);
  temperature = Math.max(temperature, 0.01);

  return temperature;
 };
+
+type CompletionsBodyType =
+  | ChatCompletionCreateParamsNonStreaming
+  | ChatCompletionCreateParamsStreaming;
+
+export const llmCompletionsBodyFormat = <T extends CompletionsBodyType>(
+  body: T,
+  model: string | LLMModelItemType
+) => {
+  const modelData = typeof model === 'string' ? getLLMModel(model) : model;
+  if (!modelData) {
+    return body;
+  }
+
+  const requestBody: T = {
+    ...body,
+    temperature: body.temperature
+      ? computedTemperature({
+          model: modelData,
+          temperature: body.temperature
+        })
+      : undefined,
+    ...modelData?.defaultConfig
+  };
+
+  // field map
+  if (modelData.fieldMap) {
+    Object.entries(modelData.fieldMap).forEach(([sourceKey, targetKey]) => {
+      // @ts-ignore
+      requestBody[targetKey] = body[sourceKey];
+      // @ts-ignore
+      delete requestBody[sourceKey];
+    });
+  }
+
+  // console.log(requestBody);
+
+  return requestBody;
+};
--- a/packages/service/core/dataset/search/controller.ts
+++ b/packages/service/core/dataset/search/controller.ts
@@ -271,7 +271,7 @@ export async function searchDatasetData(props: SearchDatasetDataProps) {
        collectionId: { $in: Array.from(new Set(results.map((item) => item.collectionId))) },
        'indexes.dataId': { $in: results.map((item) => item.id?.trim()) }
      },
-      'datasetId collectionId q a chunkIndex indexes'
+      'datasetId collectionId updateTime q a chunkIndex indexes'
    )
      .populate('collectionId', 'name fileId rawLink externalFileId externalFileUrl')
      .lean()) as DatasetDataWithCollectionType[];
@@ -299,6 +299,7 @@ export async function searchDatasetData(props: SearchDatasetDataProps) {

      const result: SearchDataResponseItemType = {
        id: String(data._id),
+        updateTime: data.updateTime,
        q: data.q,
        a: data.a,
        chunkIndex: data.chunkIndex,
@@ -396,6 +397,7 @@ export async function searchDatasetData(props: SearchDatasetDataProps) {
                _id: 1,
                datasetId: 1,
                collectionId: 1,
+                updateTime: 1,
                q: 1,
                a: 1,
                chunkIndex: 1,
@@ -425,6 +427,7 @@ export async function searchDatasetData(props: SearchDatasetDataProps) {
          id: String(item._id),
          datasetId: String(item.datasetId),
          collectionId: String(item.collectionId),
+          updateTime: item.updateTime,
          ...getCollectionSourceData(collection),
          q: item.q,
          a: item.a,
--- a/packages/service/core/workflow/dispatch/agent/classifyQuestion.ts
+++ b/packages/service/core/workflow/dispatch/agent/classifyQuestion.ts
@@ -17,6 +17,7 @@ import { DispatchNodeResultType } from '@fastgpt/global/core/workflow/runtime/ty
 import { chatValue2RuntimePrompt } from '@fastgpt/global/core/chat/adapt';
 import { getHandleId } from '@fastgpt/global/core/workflow/utils';
 import { loadRequestMessages } from '../../../chat/utils';
+import { llmCompletionsBodyFormat } from '../../../ai/utils';

 type Props = ModuleDispatchProps<{
  [NodeInputKeyEnum.aiModel]: string;
@@ -103,7 +104,7 @@ const completions = async ({
              systemPrompt: systemPrompt || 'null',
              typeList: agents
                .map((item) => `{"类型ID":"${item.key}", "问题类型":"${item.value}"}`)
-                .join('------'),
+                .join('\n------\n'),
              history: histories
                .map((item) => `${item.obj}:${chatValue2RuntimePrompt(item.value).text}`)
                .join('------'),
@@ -124,13 +125,17 @@ const completions = async ({
    timeout: 480000
  });

-  const data = await ai.chat.completions.create({
-    model: cqModel.model,
-    temperature: 0.01,
-    messages: requestMessages,
-    stream: false,
-    ...cqModel.defaultConfig
-  });
+  const data = await ai.chat.completions.create(
+    llmCompletionsBodyFormat(
+      {
+        model: cqModel.model,
+        temperature: 0.01,
+        messages: requestMessages,
+        stream: false
+      },
+      cqModel
+    )
+  );
  const answer = data.choices?.[0].message?.content || '';

  // console.log(JSON.stringify(chats2GPTMessages({ messages, reserveId: false }), null, 2));
--- a/packages/service/core/workflow/dispatch/agent/extract.ts
+++ b/packages/service/core/workflow/dispatch/agent/extract.ts
@@ -26,6 +26,7 @@ import {
 import { ChatCompletionRequestMessageRoleEnum } from '@fastgpt/global/core/ai/constants';
 import { DispatchNodeResultType } from '@fastgpt/global/core/workflow/runtime/type';
 import { chatValue2RuntimePrompt } from '@fastgpt/global/core/chat/adapt';
+import { llmCompletionsBodyFormat } from '../../../ai/utils';

 type Props = ModuleDispatchProps<{
  [NodeInputKeyEnum.history]?: ChatItemType[];
@@ -161,7 +162,7 @@ ${description ? `- ${description}` : ''}
 - 需要结合前面的对话内容，一起生成合适的参数。
 """

-本次输入内容: ${content}
+本次输入内容: """${content}"""
            `
          }
        }
@@ -226,13 +227,18 @@ const toolChoice = async (props: ActionProps) => {
    timeout: 480000
  });

-  const response = await ai.chat.completions.create({
-    model: extractModel.model,
-    temperature: 0.01,
-    messages: filterMessages,
-    tools,
-    tool_choice: { type: 'function', function: { name: agentFunName } }
-  });
+  const response = await ai.chat.completions.create(
+    llmCompletionsBodyFormat(
+      {
+        model: extractModel.model,
+        temperature: 0.01,
+        messages: filterMessages,
+        tools,
+        tool_choice: { type: 'function', function: { name: agentFunName } }
+      },
+      extractModel
+    )
+  );

  const arg: Record<string, any> = (() => {
    try {
@@ -271,15 +277,20 @@ const functionCall = async (props: ActionProps) => {
    timeout: 480000
  });

-  const response = await ai.chat.completions.create({
-    model: extractModel.model,
-    temperature: 0.01,
-    messages: filterMessages,
-    function_call: {
-      name: agentFunName
-    },
-    functions
-  });
+  const response = await ai.chat.completions.create(
+    llmCompletionsBodyFormat(
+      {
+        model: extractModel.model,
+        temperature: 0.01,
+        messages: filterMessages,
+        function_call: {
+          name: agentFunName
+        },
+        functions
+      },
+      extractModel
+    )
+  );

  try {
    const arg = JSON.parse(response?.choices?.[0]?.message?.function_call?.arguments || '');
@@ -311,7 +322,7 @@ const completions = async ({
  extractModel,
  user,
  histories,
-  params: { content, extractKeys, description }
+  params: { content, extractKeys, description = 'No special requirements' }
 }: ActionProps) => {
  const messages: ChatItemType[] = [
    {
@@ -351,13 +362,17 @@ Human: ${content}`
    userKey: user.openaiAccount,
    timeout: 480000
  });
-  const data = await ai.chat.completions.create({
-    model: extractModel.model,
-    temperature: 0.01,
-    messages: requestMessages,
-    stream: false,
-    ...extractModel.defaultConfig
-  });
+  const data = await ai.chat.completions.create(
+    llmCompletionsBodyFormat(
+      {
+        model: extractModel.model,
+        temperature: 0.01,
+        messages: requestMessages,
+        stream: false
+      },
+      extractModel
+    )
+  );
  const answer = data.choices?.[0].message?.content || '';

  // parse response
--- a/packages/service/core/workflow/dispatch/agent/runTool/functionCall.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/functionCall.ts
@@ -24,7 +24,7 @@ import { getNanoid, sliceStrStartEnd } from '@fastgpt/global/common/string/tools
 import { AIChatItemType } from '@fastgpt/global/core/chat/type';
 import { GPTMessages2Chats } from '@fastgpt/global/core/chat/adapt';
 import { updateToolInputValue } from './utils';
-import { computedMaxToken, computedTemperature } from '../../../../ai/utils';
+import { computedMaxToken, llmCompletionsBodyFormat } from '../../../../ai/utils';

 type FunctionRunResponseType = {
  toolRunResponse: DispatchFlowResponse;
@@ -110,19 +110,18 @@ export const runToolWithFunctionCall = async (
      filterMessages
    })
  ]);
-  const requestBody: any = {
-    ...toolModel?.defaultConfig,
-    model: toolModel.model,
-    temperature: computedTemperature({
-      model: toolModel,
-      temperature
-    }),
-    max_tokens,
-    stream,
-    messages: requestMessages,
-    functions,
-    function_call: 'auto'
-  };
+  const requestBody = llmCompletionsBodyFormat(
+    {
+      model: toolModel.model,
+      temperature,
+      max_tokens,
+      stream,
+      messages: requestMessages,
+      functions,
+      function_call: 'auto'
+    },
+    toolModel
+  );

  // console.log(JSON.stringify(requestBody, null, 2));
  /* Run llm */
--- a/packages/service/core/workflow/dispatch/agent/runTool/promptCall.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/promptCall.ts
@@ -25,7 +25,7 @@ import {
 import { AIChatItemType } from '@fastgpt/global/core/chat/type';
 import { GPTMessages2Chats } from '@fastgpt/global/core/chat/adapt';
 import { updateToolInputValue } from './utils';
-import { computedMaxToken, computedTemperature } from '../../../../ai/utils';
+import { computedMaxToken, llmCompletionsBodyFormat } from '../../../../ai/utils';
 import { WorkflowResponseType } from '../../type';

 type FunctionCallCompletion = {
@@ -113,18 +113,16 @@ export const runToolWithPromptCall = async (
      filterMessages
    })
  ]);
-  const requestBody = {
-    model: toolModel.model,
-    temperature: computedTemperature({
-      model: toolModel,
-      temperature
-    }),
-    max_completion_tokens: max_tokens,
-    max_tokens,
-    stream,
-    messages: requestMessages,
-    ...toolModel?.defaultConfig
-  };
+  const requestBody = llmCompletionsBodyFormat(
+    {
+      model: toolModel.model,
+      temperature,
+      max_tokens,
+      stream,
+      messages: requestMessages
+    },
+    toolModel
+  );

  // console.log(JSON.stringify(requestBody, null, 2));
  /* Run llm */
--- a/packages/service/core/workflow/dispatch/agent/runTool/toolChoice.ts
+++ b/packages/service/core/workflow/dispatch/agent/runTool/toolChoice.ts
@@ -24,7 +24,7 @@ import { countGptMessagesTokens } from '../../../../../common/string/tiktoken/in
 import { GPTMessages2Chats } from '@fastgpt/global/core/chat/adapt';
 import { AIChatItemType } from '@fastgpt/global/core/chat/type';
 import { updateToolInputValue } from './utils';
-import { computedMaxToken, computedTemperature } from '../../../../ai/utils';
+import { computedMaxToken, llmCompletionsBodyFormat } from '../../../../ai/utils';
 import { getNanoid, sliceStrStartEnd } from '@fastgpt/global/common/string/tools';
 import { addLog } from '../../../../../common/system/log';

@@ -127,20 +127,18 @@ export const runToolWithToolChoice = async (
      filterMessages
    })
  ]);
-  const requestBody: any = {
-    model: toolModel.model,
-    temperature: computedTemperature({
-      model: toolModel,
-      temperature
-    }),
-    max_completion_tokens: max_tokens,
-    max_tokens,
-    stream,
-    messages: requestMessages,
-    tools,
-    tool_choice: 'auto',
-    ...toolModel?.defaultConfig
-  };
+  const requestBody = llmCompletionsBodyFormat(
+    {
+      model: toolModel.model,
+      temperature,
+      max_tokens,
+      stream,
+      messages: requestMessages,
+      tools,
+      tool_choice: 'auto'
+    },
+    toolModel
+  );

  // console.log(JSON.stringify(requestBody, null, 2));
  /* Run llm */
--- a/packages/service/core/workflow/dispatch/chat/oneapi.ts
+++ b/packages/service/core/workflow/dispatch/chat/oneapi.ts
@@ -25,8 +25,9 @@ import {
 } from '@fastgpt/global/core/chat/adapt';
 import {
  Prompt_DocumentQuote,
-  Prompt_QuotePromptList,
-  Prompt_QuoteTemplateList
+  Prompt_userQuotePromptList,
+  Prompt_QuoteTemplateList,
+  Prompt_systemQuotePromptList
 } from '@fastgpt/global/core/ai/prompt/AIChat';
 import type { AIChatNodeProps } from '@fastgpt/global/core/workflow/runtime/type.d';
 import { replaceVariable } from '@fastgpt/global/common/string/tools';
@@ -40,8 +41,10 @@ import { getHistories } from '../utils';
 import { filterSearchResultsByMaxChars } from '../../utils';
 import { getHistoryPreview } from '@fastgpt/global/core/chat/utils';
 import { addLog } from '../../../../common/system/log';
-import { computedMaxToken, computedTemperature } from '../../../ai/utils';
+import { computedMaxToken, llmCompletionsBodyFormat } from '../../../ai/utils';
 import { WorkflowResponseType } from '../type';
+import { formatTime2YMDHM } from '@fastgpt/global/common/string/time';
+import { AiChatQuoteRoleType } from '@fastgpt/global/core/workflow/template/system/aiChat/type';

 export type ChatProps = ModuleDispatchProps<
  AIChatNodeProps & {
@@ -75,6 +78,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
      userChatInput,
      isResponseAnswerText = true,
      systemPrompt = '',
+      aiChatQuoteRole = 'system',
      quoteTemplate,
      quotePrompt,
      aiChatVision,
@@ -107,6 +111,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
      histories: chatHistories,
      useDatasetQuote: quoteQA !== undefined,
      datasetQuoteText,
+      aiChatQuoteRole,
      datasetQuotePrompt: quotePrompt,
      userChatInput,
      inputFiles,
@@ -152,18 +157,16 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
    })
  ]);

-  const requestBody = {
-    model: modelConstantsData.model,
-    temperature: computedTemperature({
-      model: modelConstantsData,
-      temperature
-    }),
-    max_completion_tokens: max_tokens,
-    max_tokens,
-    stream,
-    messages: requestMessages,
-    ...modelConstantsData?.defaultConfig
-  };
+  const requestBody = llmCompletionsBodyFormat(
+    {
+      model: modelConstantsData.model,
+      temperature,
+      max_tokens,
+      stream,
+      messages: requestMessages
+    },
+    modelConstantsData
+  );
  // console.log(JSON.stringify(requestBody, null, 2), '===');
  try {
    const ai = getAIApi({
@@ -279,6 +282,7 @@ async function filterDatasetQuote({
    return replaceVariable(quoteTemplate || Prompt_QuoteTemplateList[0].value, {
      q: item.q,
      a: item.a,
+      updateTime: formatTime2YMDHM(item.updateTime),
      source: item.sourceName,
      sourceId: String(item.sourceId || 'UnKnow'),
      index: index + 1
@@ -298,7 +302,8 @@ async function filterDatasetQuote({
  };
 }
 async function getChatMessages({
-  datasetQuotePrompt,
+  aiChatQuoteRole,
+  datasetQuotePrompt = '',
  datasetQuoteText,
  useDatasetQuote,
  histories = [],
@@ -308,26 +313,50 @@ async function getChatMessages({
  model,
  stringQuoteText
 }: {
+  // dataset quote
+  aiChatQuoteRole: AiChatQuoteRoleType; // user: replace user prompt; system: replace system prompt
  datasetQuotePrompt?: string;
  datasetQuoteText: string;
+
  useDatasetQuote: boolean;
  histories: ChatItemType[];
  systemPrompt: string;
  userChatInput: string;
  inputFiles: UserChatItemValueItemType['file'][];
  model: LLMModelItemType;
-  stringQuoteText?: string;
+  stringQuoteText?: string; // file quote
 }) {
-  const replaceInputValue = useDatasetQuote
-    ? replaceVariable(datasetQuotePrompt || Prompt_QuotePromptList[0].value, {
-        quote: datasetQuoteText,
-        question: userChatInput
-      })
-    : userChatInput;
+  // User role or prompt include question
+  const quoteRole =
+    aiChatQuoteRole === 'user' || datasetQuotePrompt.includes('{{question}}') ? 'user' : 'system';
+
+  const datasetQuotePromptTemplate = datasetQuotePrompt
+    ? datasetQuotePrompt
+    : quoteRole === 'user'
+      ? Prompt_userQuotePromptList[0].value
+      : Prompt_systemQuotePromptList[0].value;
+
+  const replaceInputValue =
+    useDatasetQuote && quoteRole === 'user'
+      ? replaceVariable(datasetQuotePromptTemplate, {
+          quote: datasetQuoteText,
+          question: userChatInput
+        })
+      : userChatInput;
+
+  const replaceSystemPrompt =
+    useDatasetQuote && quoteRole === 'system'
+      ? `${systemPrompt ? systemPrompt + '\n\n------\n\n' : ''}${replaceVariable(
+          datasetQuotePromptTemplate,
+          {
+            quote: datasetQuoteText
+          }
+        )}`
+      : systemPrompt;

  const messages: ChatItemType[] = [
-    ...getSystemPrompt_ChatItemType(systemPrompt),
-    ...(stringQuoteText
+    ...getSystemPrompt_ChatItemType(replaceSystemPrompt),
+    ...(stringQuoteText // file quote
      ? getSystemPrompt_ChatItemType(
          replaceVariable(Prompt_DocumentQuote, {
            quote: stringQuoteText
@@ -343,6 +372,7 @@ async function getChatMessages({
      })
    }
  ];
+
  const adaptMessages = chats2GPTMessages({ messages, reserveId: false });

  const filterMessages = await filterGPTMessageByMaxTokens({