diff --git a/docSite/content/zh-cn/docs/development/faq.md b/docSite/content/zh-cn/docs/development/faq.md index ee5e9ea1c..03e1d82f3 100644 --- a/docSite/content/zh-cn/docs/development/faq.md +++ b/docSite/content/zh-cn/docs/development/faq.md @@ -9,17 +9,31 @@ images: [] ## 一、错误排查方式 -遇到问题先按下面方式排查。 +可以先找找[Issue](https://github.com/labring/FastGPT/issues)，或新提 Issue，私有部署错误，务必提供详细的操作步骤、日志、截图，否则很难排查。 + +### 获取后端错误 1. `docker ps -a` 查看所有容器运行状态，检查是否全部 running，如有异常，尝试`docker logs 容器名`查看对应日志。 2. 容器都运行正常的，`docker logs 容器名` 查看报错日志 -3. 带有`requestId`的，都是 OneAPI 提示错误，大部分都是因为模型接口报错。 -4. 无法解决时，可以找找[Issue](https://github.com/labring/FastGPT/issues)，或新提 Issue，私有部署错误，务必提供详细的日志，否则很难排查。 +### 前端错误 + +前端报错时，页面会出现崩溃，并提示检查控制台日志。可以打开浏览器控制台，并查看`console`中的 log 日志。还可以点击对应 log 的超链接，会提示到具体错误文件，可以把这些详细错误信息提供，方便排查。 + +### OneAPI 错误 + +带有`requestId`的，都是 OneAPI 提示错误，大部分都是因为模型接口报错。可以参考 [OneAPI 常见错误](/docs/development/faq/#三常见的-oneapi-错误) ## 二、通用问题 +### 前端页面崩溃 + +1. 90% 情况是模型配置不正确：确保每类模型都至少有一个启用；检查模型中一些`对象`参数是否异常（数组和对象），如果为空，可以尝试给个空数组或空对象。 +2. 少部分是由于浏览器兼容问题，由于项目中包含一些高阶语法，可能低版本浏览器不兼容，可以将具体操作步骤和控制台中错误信息提供 issue。 +3. 关闭浏览器翻译功能，如果浏览器开启了翻译，可能会导致页面崩溃。 + ### 通过sealos部署的话，是否没有本地部署的一些限制？ + ![](/imgs/faq1.png) 这是索引模型的长度限制，通过任何方式部署都一样的，但不同索引模型的配置不一样，可以在后台修改参数。 diff --git a/docSite/content/zh-cn/docs/development/upgrading/4821.md b/docSite/content/zh-cn/docs/development/upgrading/4821.md index bc647f4f3..fc2ad35b3 100644 --- a/docSite/content/zh-cn/docs/development/upgrading/4821.md +++ b/docSite/content/zh-cn/docs/development/upgrading/4821.md @@ -11,11 +11,16 @@ weight: 804 ## 完整更新内容 -1. +1. 新增 - 弃用/已删除的插件提示。 2. 新增 - LLM 模型支持 top_p, response_format, json_schema 参数。 3. 新增 - Doubao1.5 模型预设。 4. 新增 - 向量模型支持归一化配置，以便适配未归一化的向量模型，例如 Doubao 的 embedding 模型。 -5. 优化 - 模型未配置时错误提示。 -6. 修复 - 简易模式，切换到其他非视觉模型时候，会强制关闭图片识别。 -7. 修复 - o1,o3 模型，在测试时候字段映射未生效导致报错。 -8. 修复 - 公众号对话空指针异常。 \ No newline at end of file +5. 新增 - AI 对话节点，支持输出思考过程结果，可用于其他节点引用。 +6. 优化 - 模型未配置时错误提示。 +7. 优化 - 适配非 Stream 模式思考输出。 +8. 优化 - 增加 TTS voice 未配置时的空指针保护。 +9. 优化 - Markdown 链接解析分割规则，改成严格匹配模式，牺牲兼容多种情况，减少误解析。 +10. 修复 - 简易模式，切换到其他非视觉模型时候，会强制关闭图片识别。 +11. 修复 - o1,o3 模型，在测试时候字段映射未生效导致报错。 +12. 修复 - 公众号对话空指针异常。 +13. 修复 - 多个音频/视频文件展示异常。 \ No newline at end of file diff --git a/docSite/content/zh-cn/docs/guide/workbench/workflow/dataset_search.md b/docSite/content/zh-cn/docs/guide/workbench/workflow/dataset_search.md index df8767de8..214e13b79 100644 --- a/docSite/content/zh-cn/docs/guide/workbench/workflow/dataset_search.md +++ b/docSite/content/zh-cn/docs/guide/workbench/workflow/dataset_search.md @@ -7,7 +7,7 @@ toc: true weight: 234 --- -知识库搜索具体参数说明，以及内部逻辑请移步：[FastGPT知识库搜索方案](/docs/course/data_search/) +知识库搜索具体参数说明，以及内部逻辑请移步：[FastGPT知识库搜索方案](/docs/guide/knowledge_base/rag/) ## 特点 @@ -27,7 +27,7 @@ weight: 234 ### 输入 - 搜索参数 -[点击查看参数介绍](/docs/course/data_search/#搜索参数) +[点击查看参数介绍](/docs/guide/knowledge_base/dataset_engine/#搜索参数) ### 输出 - 引用内容 diff --git a/packages/global/core/workflow/constants.ts b/packages/global/core/workflow/constants.ts index 270bb30c2..9eb9daf2d 100644 --- a/packages/global/core/workflow/constants.ts +++ b/packages/global/core/workflow/constants.ts @@ -33,8 +33,10 @@ export enum WorkflowIOValueTypeEnum { dynamic = 'dynamic', // plugin special type - selectApp = 'selectApp', - selectDataset = 'selectDataset' + selectDataset = 'selectDataset', + + // abandon + selectApp = 'selectApp' } export const toolValueTypeList = [ @@ -158,6 +160,10 @@ export enum NodeInputKeyEnum { datasetSearchExtensionBg = 'datasetSearchExtensionBg', collectionFilterMatch = 'collectionFilterMatch', authTmbId = 'authTmbId', + datasetDeepSearch = 'datasetDeepSearch', + datasetDeepSearchModel = 'datasetDeepSearchModel', + datasetDeepSearchMaxTimes = 'datasetDeepSearchMaxTimes', + datasetDeepSearchBg = 'datasetDeepSearchBg', // concat dataset datasetQuoteList = 'system_datasetQuoteList', diff --git a/packages/global/core/workflow/node/constant.ts b/packages/global/core/workflow/node/constant.ts index e3a90725f..11d8b395d 100644 --- a/packages/global/core/workflow/node/constant.ts +++ b/packages/global/core/workflow/node/constant.ts @@ -140,7 +140,14 @@ export enum FlowNodeTypeEnum { } // node IO value type -export const FlowValueTypeMap = { +export const FlowValueTypeMap: Record< + WorkflowIOValueTypeEnum, + { + label: string; + value: WorkflowIOValueTypeEnum; + abandon?: boolean; + } +> = { [WorkflowIOValueTypeEnum.string]: { label: 'String', value: WorkflowIOValueTypeEnum.string @@ -189,10 +196,6 @@ export const FlowValueTypeMap = { label: i18nT('common:core.workflow.Dataset quote'), value: WorkflowIOValueTypeEnum.datasetQuote }, - [WorkflowIOValueTypeEnum.selectApp]: { - label: i18nT('common:plugin.App'), - value: WorkflowIOValueTypeEnum.selectApp - }, [WorkflowIOValueTypeEnum.selectDataset]: { label: i18nT('common:core.chat.Select dataset'), value: WorkflowIOValueTypeEnum.selectDataset @@ -200,6 +203,11 @@ export const FlowValueTypeMap = { [WorkflowIOValueTypeEnum.dynamic]: { label: i18nT('common:core.workflow.dynamic_input'), value: WorkflowIOValueTypeEnum.dynamic + }, + [WorkflowIOValueTypeEnum.selectApp]: { + label: 'selectApp', + value: WorkflowIOValueTypeEnum.selectApp, + abandon: true } }; @@ -219,3 +227,6 @@ export const datasetQuoteValueDesc = `{ q: string; a: string }[]`; +export const datasetSelectValueDesc = `{ + datasetId: string; +}[]`; diff --git a/packages/global/core/workflow/runtime/type.d.ts b/packages/global/core/workflow/runtime/type.d.ts index cc6d664c9..3626052c1 100644 --- a/packages/global/core/workflow/runtime/type.d.ts +++ b/packages/global/core/workflow/runtime/type.d.ts @@ -123,6 +123,7 @@ export type DispatchNodeResponseType = { temperature?: number; maxToken?: number; quoteList?: SearchDataResponseItemType[]; + reasoningText?: string; historyPreview?: { obj: `${ChatRoleEnum}`; value: string; @@ -133,9 +134,17 @@ export type DispatchNodeResponseType = { limit?: number; searchMode?: `${DatasetSearchModeEnum}`; searchUsingReRank?: boolean; - extensionModel?: string; - extensionResult?: string; - extensionTokens?: number; + queryExtensionResult?: { + model: string; + inputTokens: number; + outputTokens: number; + query: string; + }; + deepSearchResult?: { + model: string; + inputTokens: number; + outputTokens: number; + }; // dataset concat concatLength?: number; @@ -198,6 +207,11 @@ export type DispatchNodeResponseType = { // tool params toolParamsResult?: Record; + + // abandon + extensionModel?: string; + extensionResult?: string; + extensionTokens?: number; }; export type DispatchNodeResultType = { diff --git a/packages/global/core/workflow/template/system/aiChat/index.ts b/packages/global/core/workflow/template/system/aiChat/index.ts index 5d5cc66cc..8d6aa6cd0 100644 --- a/packages/global/core/workflow/template/system/aiChat/index.ts +++ b/packages/global/core/workflow/template/system/aiChat/index.ts @@ -151,6 +151,20 @@ export const AiChatModule: FlowNodeTemplateType = { description: i18nT('common:core.module.output.description.Ai response content'), valueType: WorkflowIOValueTypeEnum.string, type: FlowNodeOutputTypeEnum.static + }, + { + id: NodeOutputKeyEnum.reasoningText, + key: NodeOutputKeyEnum.reasoningText, + required: false, + label: i18nT('workflow:reasoning_text'), + valueType: WorkflowIOValueTypeEnum.string, + type: FlowNodeOutputTypeEnum.static, + invalid: true, + invalidCondition: ({ inputs, llmModelList }) => { + const model = inputs.find((item) => item.key === NodeInputKeyEnum.aiModel)?.value; + const modelItem = llmModelList.find((item) => item.model === model); + return modelItem?.reasoning !== true; + } } ] }; diff --git a/packages/global/core/workflow/template/system/datasetSearch.ts b/packages/global/core/workflow/template/system/datasetSearch.ts index 2d117ada0..f869e165f 100644 --- a/packages/global/core/workflow/template/system/datasetSearch.ts +++ b/packages/global/core/workflow/template/system/datasetSearch.ts @@ -1,5 +1,6 @@ import { datasetQuoteValueDesc, + datasetSelectValueDesc, FlowNodeInputTypeEnum, FlowNodeOutputTypeEnum, FlowNodeTypeEnum @@ -38,7 +39,8 @@ export const DatasetSearchModule: FlowNodeTemplateType = { label: i18nT('common:core.module.input.label.Select dataset'), value: [], valueType: WorkflowIOValueTypeEnum.selectDataset, - required: true + required: true, + valueDesc: datasetSelectValueDesc }, { key: NodeInputKeyEnum.datasetSimilarity, diff --git a/packages/global/core/workflow/type/io.d.ts b/packages/global/core/workflow/type/io.d.ts index 7f6b46eb7..137571d73 100644 --- a/packages/global/core/workflow/type/io.d.ts +++ b/packages/global/core/workflow/type/io.d.ts @@ -1,3 +1,4 @@ +import { LLMModelItemType } from '../../ai/model.d'; import { LLMModelTypeEnum } from '../../ai/constants'; import { WorkflowIOValueTypeEnum, NodeInputKeyEnum, NodeOutputKeyEnum } from '../constants'; import { FlowNodeInputTypeEnum, FlowNodeOutputTypeEnum } from '../node/constant'; @@ -77,6 +78,12 @@ export type FlowNodeOutputItemType = { defaultValue?: any; required?: boolean; + invalid?: boolean; + invalidCondition?: (e: { + inputs: FlowNodeInputItemType[]; + llmModelList: LLMModelItemType[]; + }) => boolean; + // component params customFieldConfig?: CustomFieldConfigType; }; diff --git a/packages/service/core/dataset/search/agent.ts b/packages/service/core/dataset/search/agent.ts deleted file mode 100644 index c5f9cea6d..000000000 --- a/packages/service/core/dataset/search/agent.ts +++ /dev/null @@ -1,277 +0,0 @@ -import { chats2GPTMessages } from '@fastgpt/global/core/chat/adapt'; -import { ChatItemType } from '@fastgpt/global/core/chat/type'; -import { DatasetSearchModeEnum } from '@fastgpt/global/core/dataset/constants'; -import { getLLMModel } from '../../ai/model'; -import { filterGPTMessageByMaxContext } from '../../chat/utils'; -import { replaceVariable } from '@fastgpt/global/common/string/tools'; -import { createChatCompletion } from '../../ai/config'; -import { llmCompletionsBodyFormat } from '../../ai/utils'; -import { ChatCompletionMessageParam } from '@fastgpt/global/core/ai/type'; -import { SearchDataResponseItemType } from '@fastgpt/global/core/dataset/type'; -import { searchDatasetData } from './controller'; - -type SearchDatasetDataProps = { - queries: string[]; - histories: ChatItemType[]; - teamId: string; - model: string; - similarity?: number; // min distance - limit: number; // max Token limit - datasetIds: string[]; - searchMode?: `${DatasetSearchModeEnum}`; - usingReRank?: boolean; - reRankQuery: string; - - /* - { - tags: { - $and: ["str1","str2"], - $or: ["str1","str2",null] null means no tags - }, - createTime: { - $gte: 'xx', - $lte: 'xxx' - } - } - */ - collectionFilterMatch?: string; -}; - -const analyzeQuery = async ({ query, histories }: { query: string; histories: ChatItemType[] }) => { - const modelData = getLLMModel('gpt-4o-mini'); - - const systemFewShot = ` -## 知识背景 -FastGPT 是低代码AI应用构建平台，支持通过语义相似度实现精准数据检索。用户正在利用该功能开发数据检索应用。 - -## 任务目标 -基于用户历史对话和知识背景，生成多维度检索方案，确保覆盖核心语义及潜在关联维度。 - -## 工作流程 -1. 问题解构阶段 - [意图识别] 提取用户问题的核心实体和关系： - - 显性需求：直接提及的关键词 - - 隐性需求：可能涉及的关联概念 - [示例] 若问题为"推荐手机"，需考虑价格、品牌、使用场景等维度 - -2. 完整性校验阶段 - [完整性评估] 检查是否缺失核心实体和关系： - - 主语完整 - - 多实体关系准确 - [维度扩展] 检查是否需要补充： - □ 时间范围 □ 地理限定 □ 比较维度 - □ 专业术语 □ 同义词替换 □ 场景参数 - -3. 检索生成阶段 - [组合策略] 生成包含以下要素的查询序列： - ① 基础查询（核心关键词） - ② 扩展查询（核心+同义词） - ③ 场景查询（核心+场景限定词） - ④ 逆向查询（相关技术/对比对象） - -## 输出规范 -格式要求： -1. 每个查询为完整陈述句 -2. 包含至少1个核心词+1个扩展维度 -3. 按查询范围从宽到窄排序 - -禁止项： -- 使用问句形式 -- 包含解决方案描述 -- 超出话题范围的假设 - -## 执行示例 -用户问题："如何优化数据检索速度" - -查询内容： -1. FastGPT 数据检索速度优化的常用方法 -2. FastGPT 大数据量下的语义检索性能提升方案 -3. FastGPT API 响应时间的优化指标 - -## 任务开始 -`.trim(); - const filterHistories = await filterGPTMessageByMaxContext({ - messages: chats2GPTMessages({ messages: histories, reserveId: false }), - maxContext: modelData.maxContext - 1000 - }); - - const messages = [ - { - role: 'system', - content: systemFewShot - }, - ...filterHistories, - { - role: 'user', - content: query - } - ] as any; - - const { response: result } = await createChatCompletion({ - body: llmCompletionsBodyFormat( - { - stream: false, - model: modelData.model, - temperature: 0.1, - messages - }, - modelData - ) - }); - let answer = result.choices?.[0]?.message?.content || ''; - - // Extract queries from the answer by line number - const queries = answer - .split('\n') - .map((line) => { - const match = line.match(/^\d+\.\s*(.+)$/); - return match ? match[1].trim() : null; - }) - .filter(Boolean) as string[]; - - if (queries.length === 0) { - return [answer]; - } - - return queries; -}; -const checkQuery = async ({ - queries, - histories, - searchResult -}: { - queries: string[]; - histories: ChatItemType[]; - searchResult: SearchDataResponseItemType[]; -}) => { - const modelData = getLLMModel('gpt-4o-mini'); - - const systemFewShot = ` -## 知识背景 -FastGPT 是低代码AI应用构建平台，支持通过语义相似度实现精准数据检索。用户正在利用该功能开发数据检索应用。 - -## 查询结果 -${searchResult.map((item) => item.q + item.a).join('---\n---')} - -## 任务目标 -检查"检索结果"是否覆盖用户的问题，如果无法覆盖用户问题，则再次生成检索方案。 - -## 工作流程 -1. 检查检索结果是否覆盖用户的问题 -2. 如果检索结果覆盖用户问题，则直接输出："Done" -3. 如果无法覆盖用户问题，则结合用户问题和检索结果，生成进一步的检索方案，进行深度检索 - -## 输出规范 - -1. 每个查询均为完整的查询语句 -2. 通过序号来表示多个检索内容 - -## 输出示例1 -Done - -## 输出示例2 -1. 环界云计算的办公地址 -2. 环界云计算的注册地址在哪里 - -## 任务开始 -`.trim(); - const filterHistories = await filterGPTMessageByMaxContext({ - messages: chats2GPTMessages({ messages: histories, reserveId: false }), - maxContext: modelData.maxContext - 1000 - }); - - const messages = [ - { - role: 'system', - content: systemFewShot - }, - ...filterHistories, - { - role: 'user', - content: queries.join('\n') - } - ] as any; - console.log(messages); - const { response: result } = await createChatCompletion({ - body: llmCompletionsBodyFormat( - { - stream: false, - model: modelData.model, - temperature: 0.1, - messages - }, - modelData - ) - }); - let answer = result.choices?.[0]?.message?.content || ''; - console.log(answer); - if (answer.includes('Done')) { - return []; - } - - const nextQueries = answer - .split('\n') - .map((line) => { - const match = line.match(/^\d+\.\s*(.+)$/); - return match ? match[1].trim() : null; - }) - .filter(Boolean) as string[]; - - return nextQueries; -}; -export const agentSearchDatasetData = async ({ - searchRes = [], - tokens = 0, - ...props -}: SearchDatasetDataProps & { - searchRes?: SearchDataResponseItemType[]; - tokens?: number; -}) => { - const query = props.queries[0]; - - const searchResultList: SearchDataResponseItemType[] = []; - let searchQueries: string[] = []; - - // 1. agent 分析问题 - searchQueries = await analyzeQuery({ query, histories: props.histories }); - - // 2. 检索内容 + 检查 - let retryTimes = 3; - while (true) { - retryTimes--; - if (retryTimes < 0) break; - - console.log(searchQueries, '--'); - const { searchRes: searchRes2, tokens: tokens2 } = await searchDatasetData({ - ...props, - queries: searchQueries - }); - // console.log(searchRes2.map((item) => item.q)); - // deduplicate and merge search results - const uniqueResults = searchRes2.filter((item) => { - return !searchResultList.some((existingItem) => existingItem.id === item.id); - }); - searchResultList.push(...uniqueResults); - if (uniqueResults.length === 0) break; - - const checkResult = await checkQuery({ - queries: searchQueries, - histories: props.histories, - searchResult: searchRes2 - }); - - if (checkResult.length > 0) { - searchQueries = checkResult; - } else { - break; - } - } - - console.log(searchResultList.length); - return { - searchRes: searchResultList, - tokens: 0, - usingSimilarityFilter: false, - usingReRank: false - }; -}; diff --git a/packages/service/core/dataset/search/controller.ts b/packages/service/core/dataset/search/controller.ts index def8613d3..3ae4e5aef 100644 --- a/packages/service/core/dataset/search/controller.ts +++ b/packages/service/core/dataset/search/controller.ts @@ -5,7 +5,7 @@ import { } from '@fastgpt/global/core/dataset/constants'; import { recallFromVectorStore } from '../../../common/vectorStore/controller'; import { getVectorsByText } from '../../ai/embedding'; -import { getEmbeddingModel, getDefaultRerankModel } from '../../ai/model'; +import { getEmbeddingModel, getDefaultRerankModel, getLLMModel } from '../../ai/model'; import { MongoDatasetData } from '../data/schema'; import { DatasetDataTextSchemaType, @@ -24,19 +24,23 @@ import { MongoDatasetCollectionTags } from '../tag/schema'; import { readFromSecondary } from '../../../common/mongo/utils'; import { MongoDatasetDataText } from '../data/dataTextSchema'; import { ChatItemType } from '@fastgpt/global/core/chat/type'; +import { POST } from '../../../common/api/plusRequest'; +import { NodeInputKeyEnum } from '@fastgpt/global/core/workflow/constants'; +import { datasetSearchQueryExtension } from './utils'; -type SearchDatasetDataProps = { - histories?: ChatItemType[]; +export type SearchDatasetDataProps = { + histories: ChatItemType[]; teamId: string; model: string; - similarity?: number; // min distance - limit: number; // max Token limit datasetIds: string[]; - searchMode?: `${DatasetSearchModeEnum}`; - usingReRank?: boolean; reRankQuery: string; queries: string[]; + [NodeInputKeyEnum.datasetSimilarity]?: number; // min distance + [NodeInputKeyEnum.datasetMaxTokens]: number; // max Token limit + [NodeInputKeyEnum.datasetSearchMode]?: `${DatasetSearchModeEnum}`; + [NodeInputKeyEnum.datasetSearchUsingReRank]?: boolean; + /* { tags: { @@ -52,7 +56,96 @@ type SearchDatasetDataProps = { collectionFilterMatch?: string; }; -export async function searchDatasetData(props: SearchDatasetDataProps) { +export type SearchDatasetDataResponse = { + searchRes: SearchDataResponseItemType[]; + tokens: number; + searchMode: `${DatasetSearchModeEnum}`; + limit: number; + similarity: number; + usingReRank: boolean; + usingSimilarityFilter: boolean; + + queryExtensionResult?: { + model: string; + inputTokens: number; + outputTokens: number; + query: string; + }; + deepSearchResult?: { model: string; inputTokens: number; outputTokens: number }; +}; + +export const datasetDataReRank = async ({ + data, + query +}: { + data: SearchDataResponseItemType[]; + query: string; +}): Promise => { + const results = await reRankRecall({ + query, + documents: data.map((item) => ({ + id: item.id, + text: `${item.q}\n${item.a}` + })) + }); + + if (results.length === 0) { + return Promise.reject('Rerank error'); + } + + // add new score to data + const mergeResult = results + .map((item, index) => { + const target = data.find((dataItem) => dataItem.id === item.id); + if (!target) return null; + const score = item.score || 0; + + return { + ...target, + score: [{ type: SearchScoreTypeEnum.reRank, value: score, index }] + }; + }) + .filter(Boolean) as SearchDataResponseItemType[]; + + return mergeResult; +}; +export const filterDatasetDataByMaxTokens = async ( + data: SearchDataResponseItemType[], + maxTokens: number +) => { + const filterMaxTokensResult = await (async () => { + // Count tokens + const tokensScoreFilter = await Promise.all( + data.map(async (item) => ({ + ...item, + tokens: await countPromptTokens(item.q + item.a) + })) + ); + + const results: SearchDataResponseItemType[] = []; + let totalTokens = 0; + + for await (const item of tokensScoreFilter) { + totalTokens += item.tokens; + + if (totalTokens > maxTokens + 500) { + break; + } + results.push(item); + if (totalTokens > maxTokens) { + break; + } + } + + return results.length === 0 ? data.slice(0, 1) : results; + })(); + + return filterMaxTokensResult; +}; + +export async function searchDatasetData( + props: SearchDatasetDataProps +): Promise { let { teamId, reRankQuery, @@ -457,47 +550,6 @@ export async function searchDatasetData(props: SearchDatasetDataProps) { tokenLen: 0 }; }; - const reRankSearchResult = async ({ - data, - query - }: { - data: SearchDataResponseItemType[]; - query: string; - }): Promise => { - try { - const results = await reRankRecall({ - query, - documents: data.map((item) => ({ - id: item.id, - text: `${item.q}\n${item.a}` - })) - }); - - if (results.length === 0) { - usingReRank = false; - return []; - } - - // add new score to data - const mergeResult = results - .map((item, index) => { - const target = data.find((dataItem) => dataItem.id === item.id); - if (!target) return null; - const score = item.score || 0; - - return { - ...target, - score: [{ type: SearchScoreTypeEnum.reRank, value: score, index }] - }; - }) - .filter(Boolean) as SearchDataResponseItemType[]; - - return mergeResult; - } catch (error) { - usingReRank = false; - return []; - } - }; const multiQueryRecall = async ({ embeddingLimit, fullTextLimit @@ -582,10 +634,15 @@ export async function searchDatasetData(props: SearchDatasetDataProps) { set.add(str); return true; }); - return reRankSearchResult({ - query: reRankQuery, - data: filterSameDataResults - }); + try { + return datasetDataReRank({ + query: reRankQuery, + data: filterSameDataResults + }); + } catch (error) { + usingReRank = false; + return []; + } })(); // embedding recall and fullText recall rrf concat @@ -630,31 +687,7 @@ export async function searchDatasetData(props: SearchDatasetDataProps) { })(); // token filter - const filterMaxTokensResult = await (async () => { - const tokensScoreFilter = await Promise.all( - scoreFilter.map(async (item) => ({ - ...item, - tokens: await countPromptTokens(item.q + item.a) - })) - ); - - const results: SearchDataResponseItemType[] = []; - let totalTokens = 0; - - for await (const item of tokensScoreFilter) { - totalTokens += item.tokens; - - if (totalTokens > maxTokens + 500) { - break; - } - results.push(item); - if (totalTokens > maxTokens) { - break; - } - } - - return results.length === 0 ? scoreFilter.slice(0, 1) : results; - })(); + const filterMaxTokensResult = await filterDatasetDataByMaxTokens(scoreFilter, maxTokens); return { searchRes: filterMaxTokensResult, @@ -666,3 +699,53 @@ export async function searchDatasetData(props: SearchDatasetDataProps) { usingSimilarityFilter }; } + +export type DefaultSearchDatasetDataProps = SearchDatasetDataProps & { + [NodeInputKeyEnum.datasetSearchUsingExtensionQuery]?: boolean; + [NodeInputKeyEnum.datasetSearchExtensionModel]?: string; + [NodeInputKeyEnum.datasetSearchExtensionBg]?: string; +}; +export const defaultSearchDatasetData = async ({ + datasetSearchUsingExtensionQuery, + datasetSearchExtensionModel, + datasetSearchExtensionBg, + ...props +}: DefaultSearchDatasetDataProps): Promise => { + const query = props.queries[0]; + + const extensionModel = datasetSearchUsingExtensionQuery + ? getLLMModel(datasetSearchExtensionModel) + : undefined; + + const { concatQueries, rewriteQuery, aiExtensionResult } = await datasetSearchQueryExtension({ + query, + extensionModel, + extensionBg: datasetSearchExtensionBg + }); + + const result = await searchDatasetData({ + ...props, + reRankQuery: rewriteQuery, + queries: concatQueries + }); + + return { + ...result, + queryExtensionResult: aiExtensionResult + ? { + model: aiExtensionResult.model, + inputTokens: aiExtensionResult.inputTokens, + outputTokens: aiExtensionResult.outputTokens, + query: concatQueries.join('\n') + } + : undefined + }; +}; + +export type DeepRagSearchProps = SearchDatasetDataProps & { + [NodeInputKeyEnum.datasetDeepSearchModel]?: string; + [NodeInputKeyEnum.datasetDeepSearchMaxTimes]?: number; + [NodeInputKeyEnum.datasetDeepSearchBg]?: string; +}; +export const deepRagSearch = (data: DeepRagSearchProps) => + POST('/core/dataset/deepRag', data); diff --git a/packages/service/core/workflow/dispatch/chat/oneapi.ts b/packages/service/core/workflow/dispatch/chat/oneapi.ts index 5f6f5ab56..1ede02afe 100644 --- a/packages/service/core/workflow/dispatch/chat/oneapi.ts +++ b/packages/service/core/workflow/dispatch/chat/oneapi.ts @@ -106,7 +106,6 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise; export type DatasetSearchResponse = DispatchNodeResultType<{ [NodeOutputKeyEnum.datasetQuoteQA]: SearchDataResponseItemType[]; @@ -52,12 +56,17 @@ export async function dispatchDatasetSearch( usingReRank, searchMode, userChatInput, + authTmbId = false, + collectionFilterMatch, datasetSearchUsingExtensionQuery, datasetSearchExtensionModel, datasetSearchExtensionBg, - collectionFilterMatch, - authTmbId = false + + datasetDeepSearch, + datasetDeepSearchModel, + datasetDeepSearchMaxTimes, + datasetDeepSearchBg } } = props as DatasetSearchProps; @@ -85,25 +94,12 @@ export async function dispatchDatasetSearch( return emptyResult; } - // query extension - const extensionModel = datasetSearchUsingExtensionQuery - ? getLLMModel(datasetSearchExtensionModel) - : undefined; - - const [{ concatQueries, rewriteQuery, aiExtensionResult }, datasetIds] = await Promise.all([ - datasetSearchQueryExtension({ - query: userChatInput, - extensionModel, - extensionBg: datasetSearchExtensionBg, - histories: getHistories(6, histories) - }), - authTmbId - ? filterDatasetsByTmbId({ - datasetIds: datasets.map((item) => item.datasetId), - tmbId - }) - : Promise.resolve(datasets.map((item) => item.datasetId)) - ]); + const datasetIds = authTmbId + ? await filterDatasetsByTmbId({ + datasetIds: datasets.map((item) => item.datasetId), + tmbId + }) + : await Promise.resolve(datasets.map((item) => item.datasetId)); if (datasetIds.length === 0) { return emptyResult; @@ -116,15 +112,11 @@ export async function dispatchDatasetSearch( ); // start search - const { - searchRes, - tokens, - usingSimilarityFilter, - usingReRank: searchUsingReRank - } = await searchDatasetData({ + const searchData = { + histories, teamId, - reRankQuery: `${rewriteQuery}`, - queries: concatQueries, + reRankQuery: userChatInput, + queries: [userChatInput], model: vectorModel.model, similarity, limit, @@ -132,59 +124,106 @@ export async function dispatchDatasetSearch( searchMode, usingReRank: usingReRank && (await checkTeamReRankPermission(teamId)), collectionFilterMatch - }); + }; + const { + searchRes, + tokens, + usingSimilarityFilter, + usingReRank: searchUsingReRank, + queryExtensionResult, + deepSearchResult + } = datasetDeepSearch + ? await deepRagSearch({ + ...searchData, + datasetDeepSearchModel, + datasetDeepSearchMaxTimes, + datasetDeepSearchBg + }) + : await defaultSearchDatasetData({ + ...searchData, + datasetSearchUsingExtensionQuery, + datasetSearchExtensionModel, + datasetSearchExtensionBg + }); // count bill results + const nodeDispatchUsages: ChatNodeUsageType[] = []; // vector - const { totalPoints, modelName } = formatModelChars2Points({ - model: vectorModel.model, - inputTokens: tokens, - modelType: ModelTypeEnum.embedding + const { totalPoints: embeddingTotalPoints, modelName: embeddingModelName } = + formatModelChars2Points({ + model: vectorModel.model, + inputTokens: tokens, + modelType: ModelTypeEnum.embedding + }); + nodeDispatchUsages.push({ + totalPoints: embeddingTotalPoints, + moduleName: node.name, + model: embeddingModelName, + inputTokens: tokens }); + // Query extension + const { totalPoints: queryExtensionTotalPoints } = (() => { + if (queryExtensionResult) { + const { totalPoints, modelName } = formatModelChars2Points({ + model: queryExtensionResult.model, + inputTokens: queryExtensionResult.inputTokens, + outputTokens: queryExtensionResult.outputTokens, + modelType: ModelTypeEnum.llm + }); + nodeDispatchUsages.push({ + totalPoints, + moduleName: i18nT('common:core.module.template.Query extension'), + model: modelName, + inputTokens: queryExtensionResult.inputTokens, + outputTokens: queryExtensionResult.outputTokens + }); + return { + totalPoints + }; + } + return { + totalPoints: 0 + }; + })(); + // Deep search + const { totalPoints: deepSearchTotalPoints } = (() => { + if (deepSearchResult) { + const { totalPoints, modelName } = formatModelChars2Points({ + model: deepSearchResult.model, + inputTokens: deepSearchResult.inputTokens, + outputTokens: deepSearchResult.outputTokens, + modelType: ModelTypeEnum.llm + }); + nodeDispatchUsages.push({ + totalPoints, + moduleName: i18nT('common:deep_rag_search'), + model: modelName, + inputTokens: deepSearchResult.inputTokens, + outputTokens: deepSearchResult.outputTokens + }); + return { + totalPoints + }; + } + return { + totalPoints: 0 + }; + })(); + const totalPoints = embeddingTotalPoints + queryExtensionTotalPoints + deepSearchTotalPoints; + const responseData: DispatchNodeResponseType & { totalPoints: number } = { totalPoints, - query: concatQueries.join('\n'), - model: modelName, + query: userChatInput, + model: vectorModel.model, inputTokens: tokens, similarity: usingSimilarityFilter ? similarity : undefined, limit, searchMode, searchUsingReRank: searchUsingReRank, - quoteList: searchRes + quoteList: searchRes, + queryExtensionResult, + deepSearchResult }; - const nodeDispatchUsages: ChatNodeUsageType[] = [ - { - totalPoints, - moduleName: node.name, - model: modelName, - inputTokens: tokens - } - ]; - - if (aiExtensionResult) { - const { totalPoints, modelName } = formatModelChars2Points({ - model: aiExtensionResult.model, - inputTokens: aiExtensionResult.inputTokens, - outputTokens: aiExtensionResult.outputTokens, - modelType: ModelTypeEnum.llm - }); - - responseData.totalPoints += totalPoints; - responseData.inputTokens = aiExtensionResult.inputTokens; - responseData.outputTokens = aiExtensionResult.outputTokens; - responseData.extensionModel = modelName; - responseData.extensionResult = - aiExtensionResult.extensionQueries?.join('\n') || - JSON.stringify(aiExtensionResult.extensionQueries); - - nodeDispatchUsages.push({ - totalPoints, - moduleName: 'core.module.template.Query extension', - model: modelName, - inputTokens: aiExtensionResult.inputTokens, - outputTokens: aiExtensionResult.outputTokens - }); - } return { quoteQA: searchRes, diff --git a/packages/service/worker/readFile/extension/pdf.ts b/packages/service/worker/readFile/extension/pdf.ts index 67da75284..ffc9e3582 100644 --- a/packages/service/worker/readFile/extension/pdf.ts +++ b/packages/service/worker/readFile/extension/pdf.ts @@ -56,14 +56,15 @@ export const readPdfFile = async ({ buffer }: ReadRawTextByBuffer): Promise i + 1); - for await (const pageNo of pageArr) { - result += await readPDFPage(doc, pageNo); + for (let i = 0; i < pageArr.length; i++) { + result += await readPDFPage(doc, i + 1); } loadingTask.destroy(); diff --git a/packages/web/components/core/workflow/NodeInputSelect.tsx b/packages/web/components/core/workflow/NodeInputSelect.tsx index 3834dab2a..ebe1df851 100644 --- a/packages/web/components/core/workflow/NodeInputSelect.tsx +++ b/packages/web/components/core/workflow/NodeInputSelect.tsx @@ -66,12 +66,6 @@ const NodeInputSelect = ({ title: t('common:core.workflow.inputType.dynamicTargetInput') }, - { - type: FlowNodeInputTypeEnum.selectApp, - icon: FlowNodeInputMap[FlowNodeInputTypeEnum.selectApp].icon, - - title: t('common:core.workflow.inputType.Manual select') - }, { type: FlowNodeInputTypeEnum.selectLLMModel, icon: FlowNodeInputMap[FlowNodeInputTypeEnum.selectLLMModel].icon, diff --git a/packages/web/i18n/en/chat.json b/packages/web/i18n/en/chat.json index 45604e158..e790daf2b 100644 --- a/packages/web/i18n/en/chat.json +++ b/packages/web/i18n/en/chat.json @@ -37,7 +37,10 @@ "not_query": "Missing query content", "not_select_file": "No file selected", "plugins_output": "Plugin Output", + "query_extension_IO_tokens": "Problem Optimization Input/Output Tokens", + "query_extension_result": "Problem optimization results", "question_tip": "From top to bottom, the response order of each module", + "reasoning_text": "Thinking process", "response.child total points": "Sub-workflow point consumption", "response.dataset_concat_length": "Combined total", "response.node_inputs": "Node Inputs", diff --git a/packages/web/i18n/en/common.json b/packages/web/i18n/en/common.json index eb79c9104..d26a51fb0 100644 --- a/packages/web/i18n/en/common.json +++ b/packages/web/i18n/en/common.json @@ -876,6 +876,7 @@ "dataset.dataset_name": "Dataset Name", "dataset.deleteFolderTips": "Confirm to Delete This Folder and All Its Contained Datasets? Data Cannot Be Recovered After Deletion, Please Confirm!", "dataset.test.noResult": "No Search Results", + "deep_rag_search": "In-depth search", "delete_api": "Are you sure you want to delete this API key? \nAfter deletion, the key will become invalid immediately and the corresponding conversation log will not be deleted. Please confirm!", "error.Create failed": "Create failed", "error.code_error": "Verification code error", @@ -883,6 +884,7 @@ "error.inheritPermissionError": "Inherit permission Error", "error.invalid_params": "Invalid parameter", "error.missingParams": "Insufficient parameters", + "error.send_auth_code_too_frequently": "Please do not obtain verification code frequently", "error.too_many_request": "Too many request", "error.upload_file_error_filename": "{{name}} Upload Failed", "error.upload_image_error": "File upload failed", diff --git a/packages/web/i18n/en/workflow.json b/packages/web/i18n/en/workflow.json index 5f7c2c7ac..94f48a3ef 100644 --- a/packages/web/i18n/en/workflow.json +++ b/packages/web/i18n/en/workflow.json @@ -139,6 +139,7 @@ "quote_role_system_tip": "Please note that the {{question}} variable is removed from the \"Quote Template Prompt Words\"", "quote_role_user_tip": "Please pay attention to adding the {{question}} variable in the \"Quote Template Prompt Word\"", "raw_response": "Raw Response", + "reasoning_text": "Thinking text", "regex": "Regex", "reply_text": "Reply Text", "request_error": "Request Error", diff --git a/packages/web/i18n/zh-CN/chat.json b/packages/web/i18n/zh-CN/chat.json index e75ae0a21..098f1e01b 100644 --- a/packages/web/i18n/zh-CN/chat.json +++ b/packages/web/i18n/zh-CN/chat.json @@ -37,7 +37,10 @@ "not_query": "缺少查询内容", "not_select_file": "未选择文件", "plugins_output": "插件输出", + "query_extension_IO_tokens": "问题优化输入/输出 Tokens", + "query_extension_result": "问题优化结果", "question_tip": "从上到下，为各个模块的响应顺序", + "reasoning_text": "思考过程", "response.child total points": "子工作流积分消耗", "response.dataset_concat_length": "合并后总数", "response.node_inputs": "节点输入", diff --git a/packages/web/i18n/zh-CN/common.json b/packages/web/i18n/zh-CN/common.json index 3b187d741..e98c7028d 100644 --- a/packages/web/i18n/zh-CN/common.json +++ b/packages/web/i18n/zh-CN/common.json @@ -879,6 +879,7 @@ "dataset.dataset_name": "知识库名称", "dataset.deleteFolderTips": "确认删除该文件夹及其包含的所有知识库？删除后数据无法恢复，请确认！", "dataset.test.noResult": "搜索结果为空", + "deep_rag_search": "深度搜索", "delete_api": "确认删除该API密钥？删除后该密钥立即失效，对应的对话日志不会删除，请确认！", "error.Create failed": "创建失败", "error.code_error": "验证码错误", @@ -886,6 +887,7 @@ "error.inheritPermissionError": "权限继承错误", "error.invalid_params": "参数无效", "error.missingParams": "参数缺失", + "error.send_auth_code_too_frequently": "请勿频繁获取验证码", "error.too_many_request": "请求太频繁了，请稍后重试", "error.upload_file_error_filename": "{{name}} 上传失败", "error.upload_image_error": "上传文件失败", diff --git a/packages/web/i18n/zh-CN/workflow.json b/packages/web/i18n/zh-CN/workflow.json index 31ca07689..dca511d58 100644 --- a/packages/web/i18n/zh-CN/workflow.json +++ b/packages/web/i18n/zh-CN/workflow.json @@ -139,6 +139,7 @@ "quote_role_system_tip": "请注意从“引用模板提示词”中移除 {{question}} 变量", "quote_role_user_tip": "请注意在“引用模板提示词”中添加 {{question}} 变量", "raw_response": "原始响应", + "reasoning_text": "思考过程", "regex": "正则", "reply_text": "回复的文本", "request_error": "请求错误", diff --git a/packages/web/i18n/zh-Hant/chat.json b/packages/web/i18n/zh-Hant/chat.json index 2980ba08b..24f28ef78 100644 --- a/packages/web/i18n/zh-Hant/chat.json +++ b/packages/web/i18n/zh-Hant/chat.json @@ -37,7 +37,9 @@ "not_query": "缺少查詢內容", "not_select_file": "尚未選取檔案", "plugins_output": "外掛程式輸出", + "query_extension_IO_tokens": "問題優化輸入/輸出 Tokens", "question_tip": "由上至下，各個模組的回應順序", + "reasoning_text": "思考過程", "response.child total points": "子工作流程點數消耗", "response.dataset_concat_length": "合併總數", "response.node_inputs": "節點輸入", diff --git a/packages/web/i18n/zh-Hant/common.json b/packages/web/i18n/zh-Hant/common.json index 52a814e42..e0278cfc7 100644 --- a/packages/web/i18n/zh-Hant/common.json +++ b/packages/web/i18n/zh-Hant/common.json @@ -876,6 +876,7 @@ "dataset.dataset_name": "知識庫名稱", "dataset.deleteFolderTips": "確認刪除此資料夾及其包含的所有知識庫？刪除後資料無法復原，請確認！", "dataset.test.noResult": "搜尋結果為空", + "deep_rag_search": "深度搜索", "delete_api": "確認刪除此 API 金鑰？\n刪除後該金鑰將立即失效，對應的對話記錄不會被刪除，請確認！", "error.Create failed": "建立失敗", "error.code_error": "驗證碼錯誤", @@ -883,6 +884,7 @@ "error.inheritPermissionError": "繼承權限錯誤", "error.invalid_params": "參數無效", "error.missingParams": "參數不足", + "error.send_auth_code_too_frequently": "請勿頻繁獲取驗證碼", "error.too_many_request": "請求太頻繁了，請稍後重試", "error.upload_file_error_filename": "{{name}} 上傳失敗", "error.upload_image_error": "上傳文件失敗", diff --git a/packages/web/i18n/zh-Hant/workflow.json b/packages/web/i18n/zh-Hant/workflow.json index fe9e93e6c..c542addce 100644 --- a/packages/web/i18n/zh-Hant/workflow.json +++ b/packages/web/i18n/zh-Hant/workflow.json @@ -139,6 +139,7 @@ "quote_role_system_tip": "請注意從「引用範本提示詞」中移除 {{question}} 變數", "quote_role_user_tip": "請注意在「引用範本提示詞」中加入 {{question}} 變數", "raw_response": "原始回應", + "reasoning_text": "思考過程", "regex": "正規表達式", "reply_text": "回覆文字", "request_error": "請求錯誤", diff --git a/projects/app/src/components/Markdown/codeBlock/Audio.tsx b/projects/app/src/components/Markdown/codeBlock/Audio.tsx index a1510bc4d..f97d7379f 100644 --- a/projects/app/src/components/Markdown/codeBlock/Audio.tsx +++ b/projects/app/src/components/Markdown/codeBlock/Audio.tsx @@ -1,9 +1,10 @@ -import React, { useEffect } from 'react'; +import React, { useEffect, useRef } from 'react'; import { Box } from '@chakra-ui/react'; import { useMarkdownWidth } from '../hooks'; const AudioBlock = ({ code: audioUrl }: { code: string }) => { const { width, Ref } = useMarkdownWidth(); + const audioRef = useRef(null); useEffect(() => { fetch(audioUrl?.trim(), { @@ -13,8 +14,7 @@ const AudioBlock = ({ code: audioUrl }: { code: string }) => { .then((response) => response.blob()) .then((blob) => { const url = URL.createObjectURL(blob); - const audio = document.getElementById('player'); - audio?.setAttribute('src', url); + audioRef?.current?.setAttribute('src', url); }) .catch((err) => { console.log(err); @@ -22,8 +22,8 @@ const AudioBlock = ({ code: audioUrl }: { code: string }) => { }, [audioUrl]); return ( - -