Skip to content

赵月辉 / fastgpt-migrated

Go to a project
Commit fdd4e9ed by Archer Committed by GitHub
Options

Test parse cite and add tool call parallel (#4737) 

* add quote response filter (#4727)

* chatting

* add quote response filter

* add test

* remove comment

* perf: cite hidden

* perf: format llm response

* feat: comment

* update default chunk size

* update default chunk size

---------

Co-authored-by: heheer <heheer@sealos.io>
parent 683ab6c1
Showing with 577 additions and 340 deletions
docSite/content/zh-cn/docs/development/upgrading/498.md
......@@ -10,7 +10,9 @@ weight: 792
## 🚀 新增内容
1. qwen3 模型预设
1. 支持 Toolcalls 并行执行。
2. 将所有内置任务，从非 stream 模式调整成 stream 模式，避免部分模型不支持非 stream 模式。如需覆盖，则可以在模型`额外 Body`参数中，强制指定`stream=false`。
3. qwen3 模型预设
## ⚙️ 优化
......
docSite/content/zh-cn/docs/use-cases/app-cases/lab_appointment.md
......@@ -563,7 +563,7 @@ HTTP模块中，需要设置 3 个工具参数：
"hidden"
],
"label": "",
"value": 1500,
"value": 5000,
"valueType": "number"
},
{
......
packages/global/common/string/password.ts 0 → 100644
export const checkPasswordRule = (password: string) => {
const patterns = [
/\d/, // Contains digits
/[a-z]/, // Contains lowercase letters
/[A-Z]/, // Contains uppercase letters
/[!@#$%^&*()_+=-]/ // Contains special characters
];
const validChars = /^[\dA-Za-z!@#$%^&*()_+=-]{6,100}$/;
// Check length and valid characters
if (!validChars.test(password)) return false;
// Count how many patterns are satisfied
const matchCount = patterns.filter((pattern) => pattern.test(password)).length;
// Must satisfy at least 2 patterns
return matchCount >= 2;
};
packages/global/core/ai/prompt/AIChat.ts
......@@ -88,8 +88,8 @@ export const Prompt_userQuotePromptList: PromptTemplateItem[] = [
- 保持答案与 <Reference></Reference> 中描述的一致。
- 使用 Markdown 语法优化回答格式。
- 使用与问题相同的语言回答。
- 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)。"
- 使用 [id](CITE) 格式来引用<Reference></Reference>中的知识，其中 CITE 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](CITE)。"
- 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。`,
['4.9.2']: `使用 <Reference></Reference> 标记中的内容作为本次对话的参考:
......@@ -146,8 +146,8 @@ export const Prompt_userQuotePromptList: PromptTemplateItem[] = [
- 保持答案与 <Reference></Reference> 中描述的一致。
- 使用 Markdown 语法优化回答格式。
- 使用与问题相同的语言回答。
- 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)。"
- 使用 [id](CITE) 格式来引用<Reference></Reference>中的知识，其中 CITE 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](CITE)。"
- 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。
问题:"""{{question}}"""`,
......@@ -217,8 +217,8 @@ export const Prompt_systemQuotePromptList: PromptTemplateItem[] = [
- 保持答案与 <Reference></Reference> 中描述的一致。
- 使用 Markdown 语法优化回答格式。
- 使用与问题相同的语言回答。
- 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)。"
- 使用 [id](CITE) 格式来引用<Reference></Reference>中的知识，其中 CITE 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](CITE)。"
- 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。`,
['4.9.2']: `使用 <Reference></Reference> 标记中的内容作为本次对话的参考:
......@@ -271,8 +271,8 @@ export const Prompt_systemQuotePromptList: PromptTemplateItem[] = [
- 保持答案与 <Reference></Reference> 中描述的一致。
- 使用 Markdown 语法优化回答格式。
- 使用与问题相同的语言回答。
- 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)。"
- 使用 [id](CITE) 格式来引用<Reference></Reference>中的知识，其中 CITE 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](CITE)。"
- 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。
问题:"""{{question}}"""`,
......@@ -321,24 +321,13 @@ export const Prompt_systemQuotePromptList: PromptTemplateItem[] = [
}
];
export const getQuotePrompt = (
version?: string,
role: 'user' | 'system' = 'user',
parseQuote = true
) => {
export const getQuotePrompt = (version?: string, role: 'user' | 'system' = 'user') => {
const quotePromptTemplates =
role === 'user' ? Prompt_userQuotePromptList : Prompt_systemQuotePromptList;
const defaultTemplate = quotePromptTemplates[0].value;
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。`,
''
);
return getPromptByVersion(version, defaultTemplate);
};
// Document quote prompt
......
packages/global/core/ai/prompt/agent.ts
......@@ -60,7 +60,7 @@ export const getExtractJsonToolPrompt = (version?: string) => {
"""
- {{description}}
- 不是每个参数都是必须生成的，如果没有合适的参数值，不要生成该参数，或返回空字符串。
- 需要结合前面的对话内容，一起生成合适的参数。
- 需要结合历史记录，一起生成合适的参数。
"""
本次输入内容: """{{content}}"""
......
packages/global/core/ai/prompt/dataset.ts
export const getDatasetSearchToolResponsePrompt = (parseQuote: boolean) => {
return parseQuote
? `## Role
export const getDatasetSearchToolResponsePrompt = () => {
return `## 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
......@@ -9,16 +8,7 @@ export const getDatasetSearchToolResponsePrompt = (parseQuote: boolean) => {
- 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。
- 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。`
: `## 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。`;
- 使用 [id](CITE) 格式来引用 "quotes" 中的知识，其中 CITE 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](CITE)。"
- 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。`;
};
packages/global/core/ai/type.d.ts
......@@ -60,6 +60,7 @@ export type ChatCompletionAssistantToolParam = {
tool_calls: ChatCompletionMessageToolCall[];
};
export type ChatCompletionMessageToolCall = ChatCompletionMessageToolCall & {
index?: number;
toolName?: string;
toolAvatar?: string;
};
......
packages/global/core/chat/utils.ts
import { DispatchNodeResponseType } from '../workflow/runtime/type';
import { FlowNodeTypeEnum } from '../workflow/node/constant';
import { ChatItemValueTypeEnum, ChatRoleEnum, ChatSourceEnum } from './constants';
import { ChatHistoryItemResType, ChatItemType, UserChatItemValueItemType } from './type.d';
import {
AIChatItemValueItemType,
ChatHistoryItemResType,
ChatItemType,
UserChatItemValueItemType
} from './type.d';
import { sliceStrStartEnd } from '../../common/string/tools';
import { PublishChannelEnum } from '../../support/outLink/constant';
import { removeDatasetCiteText } from '../../../service/core/ai/utils';
// Concat 2 -> 1, and sort by role
export const concatHistories = (histories1: ChatItemType[], histories2: ChatItemType[]) => {
......@@ -77,6 +83,7 @@ export const getHistoryPreview = (
});
};
// Filter workflow public response
export const filterPublicNodeResponseData = ({
flowResponses = [],
responseDetail = false
......@@ -112,6 +119,40 @@ export const filterPublicNodeResponseData = ({
});
};
// Remove dataset cite in ai response
export const removeAIResponseCite = <T extends AIChatItemValueItemType[] | string>(
value: T,
retainCite: boolean
): T => {
if (retainCite) return value;
if (typeof value === 'string') {
return removeDatasetCiteText(value, false) as T;
}
return value.map<AIChatItemValueItemType>((item) => {
if (item.text?.content) {
return {
...item,
text: {
...item.text,
content: removeDatasetCiteText(item.text.content, false)
}
};
}
if (item.reasoning?.content) {
return {
...item,
reasoning: {
...item.reasoning,
content: removeDatasetCiteText(item.reasoning.content, false)
}
};
}
return item;
}) as T;
};
export const removeEmptyUserInput = (input?: UserChatItemValueItemType[]) => {
return (
input?.filter((item) => {
......
packages/global/core/dataset/training/utils.ts
......@@ -8,7 +8,7 @@ import {
export const minChunkSize = 64; // min index and chunk size
// Chunk size
export const chunkAutoChunkSize = 1500;
export const chunkAutoChunkSize = 1000;
export const getMaxChunkSize = (model: LLMModelItemType) => {
return Math.max(model.maxContext - model.maxResponse, 2000);
};
......
packages/global/core/workflow/runtime/type.d.ts
......@@ -58,7 +58,7 @@ export type ChatDispatchProps = {
chatConfig: AppSchema['chatConfig'];
lastInteractive?: WorkflowInteractiveResponseType; // last interactive response
stream: boolean;
parseQuote?: boolean;
retainDatasetCite?: boolean;
maxRunTimes: number;
isToolCall?: boolean;
workflowStreamResponse?: WorkflowResponseType;
......
packages/global/core/workflow/template/system/datasetSearch.ts
......@@ -54,7 +54,7 @@ export const DatasetSearchModule: FlowNodeTemplateType = {
key: NodeInputKeyEnum.datasetMaxTokens,
renderTypeList: [FlowNodeInputTypeEnum.hidden],
label: '',
value: 1500,
value: 5000,
valueType: WorkflowIOValueTypeEnum.number
},
{
......
packages/service/core/ai/config/provider/Qwen.json
......@@ -2,14 +2,14 @@
"provider": "Qwen",
"list": [
{
"model": "qwen-vl-plus",
"name": "qwen-vl-plus",
"model": "qwen-max",
"name": "Qwen-max",
"maxContext": 32000,
"maxResponse": 2000,
"quoteMaxToken": 20000,
"maxTemperature": 1.2,
"vision": true,
"toolChoice": false,
"maxResponse": 4000,
"quoteMaxToken": 6000,
"maxTemperature": 1,
"vision": false,
"toolChoice": true,
"functionCall": false,
"defaultSystemChatPrompt": "",
"datasetProcess": true,
......@@ -19,19 +19,22 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"fieldMap": {},
"type": "llm",
"showTopP": true,
"showStopSign": true
"showStopSign": true,
"responseFormatList": ["text", "json_object"]
},
{
"model": "qwen-plus",
"name": "Qwen-plus",
"maxContext": 64000,
"maxResponse": 8000,
"quoteMaxToken": 60000,
"maxTemperature": 1,
"vision": false,
"toolChoice": true,
"model": "qwen-vl-max",
"name": "qwen-vl-max",
"maxContext": 32000,
"maxResponse": 2000,
"quoteMaxToken": 20000,
"maxTemperature": 1.2,
"vision": true,
"toolChoice": false,
"functionCall": false,
"defaultSystemChatPrompt": "",
"datasetProcess": true,
......@@ -45,15 +48,14 @@
"fieldMap": {},
"type": "llm",
"showTopP": true,
"showStopSign": true,
"responseFormatList": ["text", "json_object"]
"showStopSign": true
},
{
"model": "qwen-turbo",
"name": "Qwen-turbo",
"maxContext": 128000,
"model": "qwen-plus",
"name": "Qwen-plus",
"maxContext": 64000,
"maxResponse": 8000,
"quoteMaxToken": 100000,
"quoteMaxToken": 60000,
"maxTemperature": 1,
"vision": false,
"toolChoice": true,
......@@ -73,16 +75,15 @@
"showStopSign": true,
"responseFormatList": ["text", "json_object"]
},
{
"model": "qwen-max",
"name": "Qwen-max",
"model": "qwen-vl-plus",
"name": "qwen-vl-plus",
"maxContext": 32000,
"maxResponse": 4000,
"quoteMaxToken": 6000,
"maxTemperature": 1,
"vision": false,
"toolChoice": true,
"maxResponse": 2000,
"quoteMaxToken": 20000,
"maxTemperature": 1.2,
"vision": true,
"toolChoice": false,
"functionCall": false,
"defaultSystemChatPrompt": "",
"datasetProcess": true,
......@@ -92,22 +93,19 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"fieldMap": {},
"type": "llm",
"showTopP": true,
"showStopSign": true,
"responseFormatList": ["text", "json_object"]
"showStopSign": true
},
{
"model": "qwen-vl-max",
"name": "qwen-vl-max",
"maxContext": 32000,
"maxResponse": 2000,
"quoteMaxToken": 20000,
"maxTemperature": 1.2,
"vision": true,
"toolChoice": false,
"model": "qwen-turbo",
"name": "Qwen-turbo",
"maxContext": 128000,
"maxResponse": 8000,
"quoteMaxToken": 100000,
"maxTemperature": 1,
"vision": false,
"toolChoice": true,
"functionCall": false,
"defaultSystemChatPrompt": "",
"datasetProcess": true,
......@@ -121,7 +119,8 @@
"fieldMap": {},
"type": "llm",
"showTopP": true,
"showStopSign": true
"showStopSign": true,
"responseFormatList": ["text", "json_object"]
},
{
"model": "qwen3-235b-a22b",
......@@ -142,7 +141,9 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": true,
......@@ -168,7 +169,9 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": true,
......@@ -194,7 +197,9 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": true,
......@@ -220,7 +225,9 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": true,
......@@ -246,7 +253,9 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": true,
......@@ -272,7 +281,9 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": true,
......@@ -298,7 +309,9 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": true,
......@@ -324,7 +337,9 @@
"usedInQueryExtension": true,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": true,
......@@ -350,7 +365,9 @@
"usedInQueryExtension": false,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": false,
......@@ -375,7 +392,9 @@
"usedInQueryExtension": false,
"customExtractPrompt": "",
"usedInToolCall": true,
"defaultConfig": {},
"defaultConfig": {
"stream": true
},
"fieldMap": {},
"type": "llm",
"showTopP": false,
......
packages/service/core/ai/functions/createQuestionGuide.ts
......@@ -2,7 +2,7 @@ import type { ChatCompletionMessageParam } from '@fastgpt/global/core/ai/type.d'
import { createChatCompletion } from '../config';
import { countGptMessagesTokens, countPromptTokens } from '../../../common/string/tiktoken/index';
import { loadRequestMessages } from '../../chat/utils';
import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '../utils';
import { llmCompletionsBodyFormat, formatLLMResponse } from '../utils';
import {
QuestionGuidePrompt,
QuestionGuideFooterPrompt
......@@ -42,12 +42,12 @@ export async function createQuestionGuide({
temperature: 0.1,
max_tokens: 200,
messages: requestMessages,
stream: false
stream: true
},
model
)
});
const { text: answer, usage } = await llmResponseToAnswerText(response);
const { text: answer, usage } = await formatLLMResponse(response);
const start = answer.indexOf('[');
const end = answer.lastIndexOf(']');
......
packages/service/core/ai/functions/queryExtension.ts
......@@ -4,7 +4,7 @@ import { ChatItemType } from '@fastgpt/global/core/chat/type';
import { countGptMessagesTokens, countPromptTokens } from '../../../common/string/tiktoken/index';
import { chats2GPTMessages } from '@fastgpt/global/core/chat/adapt';
import { getLLMModel } from '../model';
import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '../utils';
import { llmCompletionsBodyFormat, formatLLMResponse } from '../utils';
import { addLog } from '../../../common/system/log';
import { filterGPTMessageByMaxContext } from '../../chat/utils';
import json5 from 'json5';
......@@ -170,7 +170,7 @@ assistant: ${chatBg}
const { response } = await createChatCompletion({
body: llmCompletionsBodyFormat(
{
stream: false,
stream: true,
model: modelData.model,
temperature: 0.1,
messages
......@@ -178,7 +178,7 @@ assistant: ${chatBg}
modelData
)
});
const { text: answer, usage } = await llmResponseToAnswerText(response);
const { text: answer, usage } = await formatLLMResponse(response);
const inputTokens = usage?.prompt_tokens || (await countGptMessagesTokens(messages));
const outputTokens = usage?.completion_tokens || (await countPromptTokens(answer));
......
packages/service/core/app/plugin/utils.ts
......@@ -31,5 +31,6 @@ export const computedPluginUsage = async ({
return plugin.hasTokenFee ? pluginCurrentCost + childrenUsages : pluginCurrentCost;
}
// Personal plugins are charged regardless of whether they are successful or not
return childrenUsages;
};
packages/service/core/workflow/dispatch/agent/classifyQuestion.ts
......@@ -19,7 +19,7 @@ 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, llmResponseToAnswerText } from '../../../ai/utils';
import { llmCompletionsBodyFormat, formatLLMResponse } from '../../../ai/utils';
import { addLog } from '../../../../common/system/log';
import { ModelTypeEnum } from '../../../../../global/core/ai/model';
import { replaceVariable } from '@fastgpt/global/common/string/tools';
......@@ -135,13 +135,13 @@ const completions = async ({
model: cqModel.model,
temperature: 0.01,
messages: requestMessages,
stream: false
stream: true
},
cqModel
),
userKey: externalProvider.openaiAccount
});
const { text: answer, usage } = await llmResponseToAnswerText(response);
const { text: answer, usage } = await formatLLMResponse(response);
// console.log(JSON.stringify(chats2GPTMessages({ messages, reserveId: false }), null, 2));
// console.log(answer, '----');
......
packages/service/core/workflow/dispatch/agent/extract.ts
......@@ -30,7 +30,7 @@ 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, llmResponseToAnswerText } from '../../../ai/utils';
import { llmCompletionsBodyFormat, formatLLMResponse } from '../../../ai/utils';
import { ModelTypeEnum } from '../../../../../global/core/ai/model';
import {
getExtractJsonPrompt,
......@@ -226,10 +226,10 @@ const toolChoice = async (props: ActionProps) => {
}
];
const { response } = (await createChatCompletion({
const { response } = await createChatCompletion({
body: llmCompletionsBodyFormat(
{
stream: false,
stream: true,
model: extractModel.model,
temperature: 0.01,
messages: filterMessages,
......@@ -239,16 +239,15 @@ const toolChoice = async (props: ActionProps) => {
extractModel
),
userKey: externalProvider.openaiAccount
})) as { response: UnStreamChatType };
});
const { toolCalls, usage } = await formatLLMResponse(response);
const arg: Record<string, any> = (() => {
try {
return json5.parse(
response?.choices?.[0]?.message?.tool_calls?.[0]?.function?.arguments || ''
);
return json5.parse(toolCalls?.[0]?.function?.arguments || '');
} catch (error) {
console.log(agentFunction.parameters);
console.log(response.choices?.[0]?.message?.tool_calls?.[0]?.function);
console.log(toolCalls?.[0]?.function);
console.log('Your model may not support tool_call', error);
return {};
}
......@@ -257,11 +256,10 @@ const toolChoice = async (props: ActionProps) => {
const AIMessages: ChatCompletionMessageParam[] = [
{
role: ChatCompletionRequestMessageRoleEnum.Assistant,
tool_calls: response.choices?.[0]?.message?.tool_calls
tool_calls: toolCalls
}
];
const usage = response.usage;
const inputTokens = usage?.prompt_tokens || (await countGptMessagesTokens(filterMessages, tools));
const outputTokens = usage?.completion_tokens || (await countGptMessagesTokens(AIMessages));
return {
......@@ -321,13 +319,13 @@ Human: ${content}`
model: extractModel.model,
temperature: 0.01,
messages: requestMessages,
stream: false
stream: true
},
extractModel
),
userKey: externalProvider.openaiAccount
});
const { text: answer, usage } = await llmResponseToAnswerText(response);
const { text: answer, usage } = await formatLLMResponse(response);
const inputTokens = usage?.prompt_tokens || (await countMessagesTokens(messages));
const outputTokens = usage?.completion_tokens || (await countPromptTokens(answer));
......
packages/service/core/workflow/dispatch/agent/runTool/functionCall.ts
......@@ -26,7 +26,12 @@ 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 { formatToolResponse, initToolCallEdges, initToolNodes } from './utils';
import { computedMaxToken, llmCompletionsBodyFormat } from '../../../../ai/utils';
import {
computedMaxToken,
llmCompletionsBodyFormat,
removeDatasetCiteText,
parseLLMStreamResponse
} from '../../../../ai/utils';
import { toolValueTypeList } from '@fastgpt/global/core/workflow/constants';
import { WorkflowInteractiveResponseType } from '@fastgpt/global/core/workflow/template/system/interactive/type';
import { ChatItemValueTypeEnum } from '@fastgpt/global/core/chat/constants';
......@@ -48,6 +53,7 @@ export const runToolWithFunctionCall = async (
runtimeEdges,
externalProvider,
stream,
retainDatasetCite = true,
workflowStreamResponse,
params: {
temperature,
......@@ -261,7 +267,8 @@ export const runToolWithFunctionCall = async (
res,
toolNodes,
stream: aiResponse,
workflowStreamResponse
workflowStreamResponse,
retainDatasetCite
});
return {
......@@ -288,8 +295,18 @@ export const runToolWithFunctionCall = async (
]
: [];
const answer = result.choices?.[0]?.message?.content || '';
if (answer) {
workflowStreamResponse?.({
event: SseResponseEventEnum.fastAnswer,
data: textAdaptGptResponse({
text: removeDatasetCiteText(answer, retainDatasetCite)
})
});
}
return {
answer: result.choices?.[0]?.message?.content || '',
answer,
functionCalls: toolCalls,
inputTokens: usage?.prompt_tokens,
outputTokens: usage?.completion_tokens
......@@ -509,12 +526,14 @@ async function streamResponse({
res,
toolNodes,
stream,
workflowStreamResponse
workflowStreamResponse,
retainDatasetCite
}: {
res: NextApiResponse;
toolNodes: ToolNodeItemType[];
stream: StreamChatType;
workflowStreamResponse?: WorkflowResponseType;
retainDatasetCite?: boolean;
}) {
const write = responseWriteController({
res,
......@@ -526,6 +545,8 @@ async function streamResponse({
let functionId = getNanoid();
let usage = getLLMDefaultUsage();
const { parsePart } = parseLLMStreamResponse();
for await (const part of stream) {
usage = part.usage || usage;
if (res.closed) {
......@@ -533,17 +554,21 @@ async function streamResponse({
break;
}
const responseChoice = part.choices?.[0]?.delta;
const { content: toolChoiceContent, responseContent } = parsePart({
part,
parseThinkTag: false,
retainDatasetCite
});
if (responseChoice.content) {
const content = responseChoice?.content || '';
textAnswer += content;
const responseChoice = part.choices?.[0]?.delta;
textAnswer += toolChoiceContent;
if (responseContent) {
workflowStreamResponse?.({
write,
event: SseResponseEventEnum.answer,
data: textAdaptGptResponse({
text: content
text: responseContent
})
});
} else if (responseChoice.function_call) {
......
packages/service/core/workflow/dispatch/agent/runTool/promptCall.ts
......@@ -29,8 +29,9 @@ import { formatToolResponse, initToolCallEdges, initToolNodes } from './utils';
import {
computedMaxToken,
llmCompletionsBodyFormat,
removeDatasetCiteText,
parseReasoningContent,
parseReasoningStreamContent
parseLLMStreamResponse
} from '../../../../ai/utils';
import { WorkflowResponseType } from '../../type';
import { toolValueTypeList } from '@fastgpt/global/core/workflow/constants';
......@@ -60,6 +61,7 @@ export const runToolWithPromptCall = async (
runtimeEdges,
externalProvider,
stream,
retainDatasetCite = true,
workflowStreamResponse,
params: {
temperature,
......@@ -275,7 +277,8 @@ export const runToolWithPromptCall = async (
toolNodes,
stream: aiResponse,
workflowStreamResponse,
aiChatReasoning
aiChatReasoning,
retainDatasetCite
});
return {
......@@ -318,7 +321,7 @@ export const runToolWithPromptCall = async (
workflowStreamResponse?.({
event: SseResponseEventEnum.fastAnswer,
data: textAdaptGptResponse({
reasoning_content: reasoning
reasoning_content: removeDatasetCiteText(reasoning, retainDatasetCite)
})
});
}
......@@ -344,7 +347,7 @@ export const runToolWithPromptCall = async (
workflowStreamResponse?.({
event: SseResponseEventEnum.fastAnswer,
data: textAdaptGptResponse({
text: replaceAnswer
text: removeDatasetCiteText(replaceAnswer, retainDatasetCite)
})
});
}
......@@ -566,13 +569,15 @@ async function streamResponse({
res,
stream,
workflowStreamResponse,
aiChatReasoning
aiChatReasoning,
retainDatasetCite
}: {
res: NextApiResponse;
toolNodes: ToolNodeItemType[];
stream: StreamChatType;
workflowStreamResponse?: WorkflowResponseType;
aiChatReasoning?: boolean;
retainDatasetCite?: boolean;
}) {
const write = responseWriteController({
res,
......@@ -585,7 +590,7 @@ async function streamResponse({
let finish_reason: CompletionFinishReason = null;
let usage = getLLMDefaultUsage();
const { parsePart, getStartTagBuffer } = parseReasoningStreamContent();
const { parsePart } = parseLLMStreamResponse();
for await (const part of stream) {
usage = part.usage || usage;
......@@ -595,11 +600,16 @@ async function streamResponse({
break;
}
const { reasoningContent, content, finishReason } = parsePart(part, aiChatReasoning);
const { reasoningContent, content, responseContent, finishReason } = parsePart({
part,
parseThinkTag: aiChatReasoning,
retainDatasetCite
});
finish_reason = finish_reason || finishReason;
answer += content;
reasoning += reasoningContent;
// Reasoning response
if (aiChatReasoning && reasoningContent) {
workflowStreamResponse?.({
write,
......@@ -612,13 +622,15 @@ async function streamResponse({
if (content) {
if (startResponseWrite) {
workflowStreamResponse?.({
write,
event: SseResponseEventEnum.answer,
data: textAdaptGptResponse({
text: content
})
});
if (responseContent) {
workflowStreamResponse?.({
write,
event: SseResponseEventEnum.answer,
data: textAdaptGptResponse({
text: responseContent
})
});
}
} else if (answer.length >= 3) {
answer = answer.trimStart();
if (/0(:|：)/.test(answer)) {
......@@ -640,22 +652,6 @@ async function streamResponse({
}
}
if (answer === '') {
answer = getStartTagBuffer();
if (/0(:|：)/.test(answer)) {
// find first : index
const firstIndex = answer.indexOf('0:') !== -1 ? answer.indexOf('0:') : answer.indexOf('0：');
answer = answer.substring(firstIndex + 2).trim();
workflowStreamResponse?.({
write,
event: SseResponseEventEnum.answer,
data: textAdaptGptResponse({
text: answer
})
});
}
}
return { answer, reasoning, finish_reason, usage };
}
......
packages/service/core/workflow/dispatch/chat/oneapi.ts
......@@ -4,7 +4,11 @@ import type { ChatItemType, UserChatItemValueItemType } from '@fastgpt/global/co
import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
import { SseResponseEventEnum } from '@fastgpt/global/core/workflow/runtime/constants';
import { textAdaptGptResponse } from '@fastgpt/global/core/workflow/runtime/utils';
import { parseReasoningContent, parseReasoningStreamContent } from '../../../ai/utils';
import {
removeDatasetCiteText,
parseReasoningContent,
parseLLMStreamResponse
} from '../../../ai/utils';
import { createChatCompletion } from '../../../ai/config';
import type {
ChatCompletionMessageParam,
......@@ -75,7 +79,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
res,
requestOrigin,
stream = false,
parseQuote = true,
retainDatasetCite = true,
externalProvider,
histories,
node: { name, version },
......@@ -159,8 +163,7 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
userChatInput,
systemPrompt,
userFiles,
documentQuoteText,
parseQuote
documentQuoteText
}),
// Censor = true and system key, will check content
(() => {
......@@ -223,7 +226,8 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
aiChatReasoning,
parseThinkTag: modelConstantsData.reasoning,
isResponseAnswerText,
workflowStreamResponse
workflowStreamResponse,
retainDatasetCite
});
return {
......@@ -258,23 +262,21 @@ export const dispatchChatCompletion = async (props: ChatProps): Promise<ChatResp
})();
// Some models do not support streaming
if (stream) {
if (aiChatReasoning && reasoningContent) {
workflowStreamResponse?.({
event: SseResponseEventEnum.fastAnswer,
data: textAdaptGptResponse({
reasoning_content: reasoningContent
})
});
}
if (isResponseAnswerText && content) {
workflowStreamResponse?.({
event: SseResponseEventEnum.fastAnswer,
data: textAdaptGptResponse({
text: content
})
});
}
if (aiChatReasoning && reasoningContent) {
workflowStreamResponse?.({
event: SseResponseEventEnum.fastAnswer,
data: textAdaptGptResponse({
reasoning_content: removeDatasetCiteText(reasoningContent, retainDatasetCite)
})
});
}
if (isResponseAnswerText && content) {
workflowStreamResponse?.({
event: SseResponseEventEnum.fastAnswer,
data: textAdaptGptResponse({
text: removeDatasetCiteText(content, retainDatasetCite)
})
});
}
return {
......@@ -452,8 +454,7 @@ async function getChatMessages({
systemPrompt,
userChatInput,
userFiles,
documentQuoteText,
parseQuote = true
documentQuoteText
}: {
model: LLMModelItemType;
maxTokens?: number;
......@@ -470,14 +471,13 @@ 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 defaultQuotePrompt = getQuotePrompt(version, quoteRole, parseQuote);
const defaultQuotePrompt = getQuotePrompt(version, quoteRole);
const datasetQuotePromptTemplate = datasetQuotePrompt || defaultQuotePrompt;
......@@ -539,7 +539,8 @@ async function streamResponse({
workflowStreamResponse,
aiChatReasoning,
parseThinkTag,
isResponseAnswerText
isResponseAnswerText,
retainDatasetCite = true
}: {
res: NextApiResponse;
stream: StreamChatType;
......@@ -547,6 +548,7 @@ async function streamResponse({
aiChatReasoning?: boolean;
parseThinkTag?: boolean;
isResponseAnswerText?: boolean;
retainDatasetCite: boolean;
}) {
const write = responseWriteController({
res,
......@@ -557,7 +559,7 @@ async function streamResponse({
let finish_reason: CompletionFinishReason = null;
let usage: CompletionUsage = getLLMDefaultUsage();
const { parsePart, getStartTagBuffer } = parseReasoningStreamContent();
const { parsePart } = parseLLMStreamResponse();
for await (const part of stream) {
usage = part.usage || usage;
......@@ -568,7 +570,11 @@ async function streamResponse({
break;
}
const { reasoningContent, content, finishReason } = parsePart(part, parseThinkTag);
const { reasoningContent, content, responseContent, finishReason } = parsePart({
part,
parseThinkTag,
retainDatasetCite
});
finish_reason = finish_reason || finishReason;
answer += content;
reasoning += reasoningContent;
......@@ -583,26 +589,12 @@ async function streamResponse({
});
}
if (isResponseAnswerText && content) {
workflowStreamResponse?.({
write,
event: SseResponseEventEnum.answer,
data: textAdaptGptResponse({
text: content
})
});
}
}
// if answer is empty, try to get value from startTagBuffer. (Cause: The response content is too short to exceed the minimum parse length)
if (answer === '') {
answer = getStartTagBuffer();
if (isResponseAnswerText && answer) {
if (isResponseAnswerText && responseContent) {
workflowStreamResponse?.({
write,
event: SseResponseEventEnum.answer,
data: textAdaptGptResponse({
text: answer
text: responseContent
})
});
}
......
packages/service/core/workflow/dispatch/dataset/concat.ts
......@@ -21,7 +21,7 @@ export async function dispatchDatasetConcat(
props: DatasetConcatProps
): Promise<DatasetConcatResponse> {
const {
params: { limit = 1500, ...quoteMap }
params: { limit = 6000, ...quoteMap }
} = props as DatasetConcatProps;
const quoteList = Object.values(quoteMap).filter((list) => Array.isArray(list));
......
packages/service/core/workflow/dispatch/dataset/search.ts
......@@ -55,11 +55,10 @@ export async function dispatchDatasetSearch(
runningUserInfo: { tmbId },
histories,
node,
parseQuote = true,
params: {
datasets = [],
similarity,
limit = 1500,
limit = 5000,
userChatInput = '',
authTmbId = false,
collectionFilterMatch,
......@@ -114,7 +113,6 @@ export async function dispatchDatasetSearch(
if (datasetIds.length === 0) {
return emptyResult;
}
// console.log(concatQueries, rewriteQuery, aiExtensionResult);
// get vector
const vectorModel = getEmbeddingModel(
......@@ -267,7 +265,7 @@ export async function dispatchDatasetSearch(
[DispatchNodeResponseKeyEnum.nodeResponse]: responseData,
nodeDispatchUsages,
[DispatchNodeResponseKeyEnum.toolResponses]: {
prompt: getDatasetSearchToolResponsePrompt(parseQuote),
prompt: getDatasetSearchToolResponsePrompt(),
quotes: searchRes.map((item) => ({
id: item.id,
sourceName: item.sourceName,
......
packages/service/core/workflow/dispatch/index.ts
......@@ -135,7 +135,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
timezone,
externalProvider,
stream = false,
parseQuote = true,
retainDatasetCite = true,
version = 'v1',
responseDetail = true,
responseAllData = true,
......@@ -607,7 +607,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
timezone,
externalProvider,
stream,
parseQuote,
retainDatasetCite,
node,
runtimeNodes,
runtimeEdges,
......
packages/service/core/workflow/dispatch/utils.ts
import { getErrText } from '@fastgpt/global/common/error/utils';
import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
import type { ChatItemType } from '@fastgpt/global/core/chat/type.d';
import {
WorkflowIOValueTypeEnum,
NodeOutputKeyEnum
} from '@fastgpt/global/core/workflow/constants';
import { NodeOutputKeyEnum } from '@fastgpt/global/core/workflow/constants';
import {
RuntimeEdgeItemType,
RuntimeNodeItemType,
......@@ -34,31 +31,22 @@ export const getWorkflowResponseWrite = ({
return ({
write,
event,
data,
stream
data
}: {
write?: (text: string) => void;
event: SseResponseEventEnum;
data: Record<string, any>;
stream?: boolean; // Focus set stream response
}) => {
const useStreamResponse = stream ?? streamResponse;
const useStreamResponse = streamResponse;
if (!res || res.closed || !useStreamResponse) return;
// Forbid show detail
const detailEvent: Record<string, 1> = {
[SseResponseEventEnum.error]: 1,
[SseResponseEventEnum.flowNodeStatus]: 1,
[SseResponseEventEnum.flowResponses]: 1,
[SseResponseEventEnum.interactive]: 1,
[SseResponseEventEnum.toolCall]: 1,
[SseResponseEventEnum.toolParams]: 1,
[SseResponseEventEnum.toolResponse]: 1,
[SseResponseEventEnum.updateVariables]: 1,
[SseResponseEventEnum.flowNodeResponse]: 1
const notDetailEvent: Record<string, 1> = {
[SseResponseEventEnum.answer]: 1,
[SseResponseEventEnum.fastAnswer]: 1
};
if (!detail && detailEvent[event]) return;
if (!detail && !notDetailEvent[event]) return;
// Forbid show running status
const statusEvent: Record<string, 1> = {
......
packages/templates/src/CQ/template.json
......@@ -308,7 +308,7 @@
"key": "limit",
"renderTypeList": ["hidden"],
"label": "",
"value": 1500,
"value": 5000,
"valueType": "number"
},
{
......
packages/templates/src/simpleDatasetChat/template.json
......@@ -211,7 +211,7 @@
"key": "limit",
"renderTypeList": ["hidden"],
"label": "",
"value": 1500,
"value": 5000,
"valueType": "number"
},
{
......
projects/app/src/components/Markdown/A.tsx
......@@ -21,16 +21,16 @@ import MyBox from '@fastgpt/web/components/common/MyBox';
import { getCollectionSourceData } from '@fastgpt/global/core/dataset/collection/utils';
import Markdown from '.';
import { getSourceNameIcon } from '@fastgpt/global/core/dataset/utils';
import { Types } from 'mongoose';
const A = ({ children, chatAuthData, ...props }: any) => {
const A = ({ children, chatAuthData, showAnimation, ...props }: any) => {
const { t } = useTranslation();
const { isOpen, onOpen, onClose } = useDisclosure();
const content = useMemo(() => String(children), [children]);
// empty href link
if (!props.href && typeof children?.[0] === 'string') {
const text = useMemo(() => String(children), [children]);
return (
<MyTooltip label={t('common:core.chat.markdown.Quick Question')}>
<Button
......@@ -38,16 +38,23 @@ const A = ({ children, chatAuthData, ...props }: any) => {
size={'xs'}
borderRadius={'md'}
my={1}
onClick={() => eventBus.emit(EventNameEnum.sendQuestion, { text })}
onClick={() => eventBus.emit(EventNameEnum.sendQuestion, { text: content })}
>
{text}
{content}
</Button>
</MyTooltip>
);
}
// Quote
if (props.href?.startsWith('QUOTE') && typeof children?.[0] === 'string') {
// Cite
if (
(props.href?.startsWith('CITE') || props.href?.startsWith('QUOTE')) &&
typeof children?.[0] === 'string'
) {
if (!Types.ObjectId.isValid(content)) {
return <></>;
}
const {
data: quoteData,
loading,
......@@ -74,6 +81,7 @@ const A = ({ children, chatAuthData, ...props }: any) => {
onClose={onClose}
onOpen={() => {
onOpen();
if (showAnimation) return;
getQuoteDataById(String(children));
}}
trigger={'hover'}
......@@ -90,7 +98,7 @@ const A = ({ children, chatAuthData, ...props }: any) => {
</Button>
</PopoverTrigger>
<PopoverContent boxShadow={'lg'} w={'500px'} maxW={'90vw'} py={4}>
<MyBox isLoading={loading}>
<MyBox isLoading={loading || showAnimation}>
<PopoverArrow />
<PopoverBody py={0} px={0} fontSize={'sm'}>
<Flex px={4} pb={1} justifyContent={'space-between'}>
......
projects/app/src/components/Markdown/index.tsx
......@@ -60,9 +60,9 @@ const MarkdownRender = ({
img: Image,
pre: RewritePre,
code: Code,
a: (props: any) => <A {...props} chatAuthData={chatAuthData} />
a: (props: any) => <A {...props} showAnimation={showAnimation} chatAuthData={chatAuthData} />
};
}, [chatAuthData]);
}, [chatAuthData, showAnimation]);
const formatSource = useMemo(() => {
if (showAnimation || forbidZhFormat) return source;
......
projects/app/src/components/Markdown/utils.ts
......@@ -27,14 +27,14 @@ export const mdTextFormat = (text: string) => {
return match;
});
// 处理 [quote:id] 格式引用，将 [quote:675934a198f46329dfc6d05a] 转换为 [675934a198f46329dfc6d05a](QUOTE)
// 处理 [quote:id] 格式引用，将 [quote:675934a198f46329dfc6d05a] 转换为 [675934a198f46329dfc6d05a](CITE)
text = text
// .replace(
// /([\u4e00-\u9fa5\u3000-\u303f])([a-zA-Z0-9])|([a-zA-Z0-9])([\u4e00-\u9fa5\u3000-\u303f])/g,
// '$1$3 $2$4'
// )
// 处理 格式引用，将 [675934a198f46329dfc6d05a] 转换为 [675934a198f46329dfc6d05a](QUOTE)
.replace(/\[([a-f0-9]{24})\](?!\()/g, '[$1](QUOTE)');
// 处理 格式引用，将 [675934a198f46329dfc6d05a] 转换为 [675934a198f46329dfc6d05a](CITE)
.replace(/\[([a-f0-9]{24})\](?!\()/g, '[$1](CITE)');
// 处理链接后的中文标点符号，增加空格
text = text.replace(/(https?:\/\/[^\s，。！？；：、]+)([，。！？；：、])/g, '$1 $2');
......
projects/app/src/components/core/chat/ChatContainer/ChatBox/components/ChatItem.tsx
......@@ -240,11 +240,6 @@ const ChatItem = (props: Props) => {
quoteId?: string;
}) => {
if (!setQuoteData) return;
if (isChatting)
return toast({
title: t('chat:chat.waiting_for_response'),
status: 'info'
});
const collectionIdList = collectionId
? [collectionId]
......@@ -277,18 +272,7 @@ const ChatItem = (props: Props) => {
}
});
},
[
setQuoteData,
isChatting,
toast,
t,
quoteList,
isShowReadRawSource,
appId,
chatId,
chat.dataId,
outLinkAuthData
]
[setQuoteData, quoteList, isShowReadRawSource, appId, chatId, chat.dataId, outLinkAuthData]
);
useEffect(() => {
......
projects/app/src/components/core/chat/components/AIResponseBox.tsx
......@@ -96,8 +96,6 @@ const RenderText = React.memo(function RenderText({
text: string;
chatItemDataId: string;
}) {
const isResponseDetail = useContextSelector(ChatItemContext, (v) => v.isResponseDetail);
const appId = useContextSelector(ChatBoxContext, (v) => v.appId);
const chatId = useContextSelector(ChatBoxContext, (v) => v.chatId);
const outLinkAuthData = useContextSelector(ChatBoxContext, (v) => v.outLinkAuthData);
......@@ -106,10 +104,8 @@ const RenderText = React.memo(function RenderText({
if (!text) return '';
// Remove quote references if not showing response detail
return isResponseDetail
? text
: text.replace(/\[([a-f0-9]{24})\]\(QUOTE\)/g, '').replace(/\[([a-f0-9]{24})\](?!\()/g, '');
}, [text, isResponseDetail]);
return text;
}, [text]);
const chatAuthData = useCreation(() => {
return { appId, chatId, chatItemDataId, ...outLinkAuthData };
......
projects/app/src/components/core/dataset/SearchParamsTip.tsx
......@@ -12,7 +12,7 @@ import { getWebLLMModel } from '@/web/common/system/utils';
const SearchParamsTip = ({
searchMode,
similarity = 0,
limit = 1500,
limit = 5000,
responseEmptyText,
usingReRank = false,
datasetSearchUsingExtensionQuery,
......
projects/app/src/pageComponents/account/info/UpdatePswModal.tsx
......@@ -5,8 +5,8 @@ import { useTranslation } from 'next-i18next';
import { useForm } from 'react-hook-form';
import { useRequest2 } from '@fastgpt/web/hooks/useRequest';
import { updatePasswordByOld } from '@/web/support/user/api';
import { checkPasswordRule } from '@/web/support/user/login/constants';
import { useToast } from '@fastgpt/web/hooks/useToast';
import { checkPasswordRule } from '@fastgpt/global/common/string/password';
type FormType = {
oldPsw: string;
......
projects/app/src/pageComponents/login/ForgetPasswordForm.tsx
import React, { Dispatch } from 'react';
import { FormControl, Box, Input, Button } from '@chakra-ui/react';
import { useForm } from 'react-hook-form';
import { LoginPageTypeEnum, checkPasswordRule } from '@/web/support/user/login/constants';
import { LoginPageTypeEnum } from '@/web/support/user/login/constants';
import { postFindPassword } from '@/web/support/user/api';
import { useSendCode } from '@/web/support/user/hooks/useSendCode';
import type { ResLogin } from '@/global/support/api/userRes.d';
......@@ -9,6 +9,7 @@ import { useToast } from '@fastgpt/web/hooks/useToast';
import { useSystemStore } from '@/web/common/system/useSystemStore';
import { useTranslation } from 'next-i18next';
import { useRequest2 } from '@fastgpt/web/hooks/useRequest';
import { checkPasswordRule } from '@fastgpt/global/common/string/password';
interface Props {
setPageType: Dispatch<`${LoginPageTypeEnum}`>;
......
projects/app/src/pageComponents/login/RegisterForm.tsx
import React, { Dispatch } from 'react';
import { FormControl, Box, Input, Button } from '@chakra-ui/react';
import { useForm } from 'react-hook-form';
import { LoginPageTypeEnum, checkPasswordRule } from '@/web/support/user/login/constants';
import { LoginPageTypeEnum } from '@/web/support/user/login/constants';
import { postRegister } from '@/web/support/user/api';
import { useSendCode } from '@/web/support/user/hooks/useSendCode';
import type { ResLogin } from '@/global/support/api/userRes';
......@@ -19,6 +19,7 @@ import {
getSourceDomain,
removeFastGPTSem
} from '@/web/support/marketing/utils';
import { checkPasswordRule } from '@fastgpt/global/common/string/password';
interface Props {
loginSuccess: (e: ResLogin) => void;
......
projects/app/src/pages/api/core/ai/model/test.ts
......@@ -16,7 +16,7 @@ import { reRankRecall } from '@fastgpt/service/core/ai/rerank';
import { aiTranscriptions } from '@fastgpt/service/core/ai/audio/transcriptions';
import { isProduction } from '@fastgpt/global/common/system/constants';
import * as fs from 'fs';
import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '@fastgpt/service/core/ai/utils';
import { llmCompletionsBodyFormat, formatLLMResponse } from '@fastgpt/service/core/ai/utils';
export type testQuery = { model: string; channelId?: number };
......@@ -78,7 +78,7 @@ const testLLMModel = async (model: LLMModelItemType, headers: Record<string, str
model
);
const { response, isStreamResponse } = await createChatCompletion({
const { response } = await createChatCompletion({
modelData: model,
body: requestBody,
options: {
......@@ -88,7 +88,7 @@ const testLLMModel = async (model: LLMModelItemType, headers: Record<string, str
}
}
});
const { text: answer } = await llmResponseToAnswerText(response);
const { text: answer } = await formatLLMResponse(response);
if (answer) {
return answer;
......
projects/app/src/pages/api/core/chat/getPaginationRecords.ts
......@@ -9,7 +9,10 @@ import { authChatCrud } from '@/service/support/permission/auth/chat';
import { MongoApp } from '@fastgpt/service/core/app/schema';
import { AppErrEnum } from '@fastgpt/global/common/error/code/app';
import { ChatItemValueTypeEnum, ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
import { filterPublicNodeResponseData } from '@fastgpt/global/core/chat/utils';
import {
filterPublicNodeResponseData,
removeAIResponseCite
} from '@fastgpt/global/core/chat/utils';
import { GetChatTypeEnum } from '@/global/core/chat/constants';
import { PaginationProps, PaginationResponse } from '@fastgpt/web/common/fetch/type';
import { ChatItemType } from '@fastgpt/global/core/chat/type';
......@@ -83,6 +86,13 @@ async function handler(
}
});
}
if (!responseDetail) {
histories.forEach((item) => {
if (item.obj === ChatRoleEnum.AI) {
item.value = removeAIResponseCite(item.value, false);
}
});
}
return {
list: isPlugin ? histories : transformPreviewHistories(histories, responseDetail),
......
projects/app/src/pages/api/core/dataset/searchTest.ts
......@@ -19,7 +19,7 @@ async function handler(req: ApiRequestProps<SearchTestProps>): Promise<SearchTes
const {
datasetId,
text,
limit = 1500,
limit = 5000,
similarity,
searchMode,
embeddingWeight,
......
projects/app/src/pages/api/v1/chat/completions.ts
......@@ -30,6 +30,7 @@ import {
concatHistories,
filterPublicNodeResponseData,
getChatTitleFromChatMessage,
removeAIResponseCite,
removeEmptyUserInput
} from '@fastgpt/global/core/chat/utils';
import { updateApiKeyUsage } from '@fastgpt/service/support/openapi/tools';
......@@ -74,7 +75,7 @@ export type Props = ChatCompletionCreateParams &
responseChatItemId?: string;
stream?: boolean;
detail?: boolean;
parseQuote?: boolean;
retainDatasetCite?: boolean;
variables: Record<string, any>; // Global variables or plugin inputs
};
......@@ -107,7 +108,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
stream = false,
detail = false,
parseQuote = false,
retainDatasetCite = false,
messages = [],
variables = {},
responseChatItemId = getNanoid(),
......@@ -187,6 +188,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
chatId
});
})();
retainDatasetCite = retainDatasetCite && !!responseDetail;
const isPlugin = app.type === AppTypeEnum.plugin;
// Check message type
......@@ -291,7 +293,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
chatConfig,
histories: newHistories,
stream,
parseQuote,
retainDatasetCite,
maxRunTimes: WORKFLOW_MAX_RUN_TIMES,
workflowStreamResponse: workflowResponseWrite
});
......@@ -406,17 +408,18 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
return assistantResponses;
})();
const formatResponseContent = removeAIResponseCite(responseContent, retainDatasetCite);
const error = flowResponses[flowResponses.length - 1]?.error;
res.json({
...(detail ? { responseData: feResponseData, newVariables } : {}),
error,
id: chatId || '',
id: saveChatId,
model: '',
usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 1 },
choices: [
{
message: { role: 'assistant', content: responseContent },
message: { role: 'assistant', content: formatResponseContent },
finish_reason: 'stop',
index: 0
}
......
projects/app/src/pages/api/v2/chat/completions.ts
......@@ -30,6 +30,7 @@ import {
concatHistories,
filterPublicNodeResponseData,
getChatTitleFromChatMessage,
removeAIResponseCite,
removeEmptyUserInput
} from '@fastgpt/global/core/chat/utils';
import { updateApiKeyUsage } from '@fastgpt/service/support/openapi/tools';
......@@ -74,7 +75,7 @@ export type Props = ChatCompletionCreateParams &
responseChatItemId?: string;
stream?: boolean;
detail?: boolean;
parseQuote?: boolean;
retainDatasetCite?: boolean;
variables: Record<string, any>; // Global variables or plugin inputs
};
......@@ -107,7 +108,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
stream = false,
detail = false,
parseQuote = false,
retainDatasetCite = false,
messages = [],
variables = {},
responseChatItemId = getNanoid(),
......@@ -187,6 +188,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
chatId
});
})();
retainDatasetCite = retainDatasetCite && !!responseDetail;
const isPlugin = app.type === AppTypeEnum.plugin;
// Check message type
......@@ -290,7 +292,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
chatConfig,
histories: newHistories,
stream,
parseQuote,
retainDatasetCite,
maxRunTimes: WORKFLOW_MAX_RUN_TIMES,
workflowStreamResponse: workflowResponseWrite,
version: 'v2',
......@@ -401,6 +403,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
return assistantResponses;
})();
const formatResponseContent = removeAIResponseCite(responseContent, retainDatasetCite);
const error = flowResponses[flowResponses.length - 1]?.error;
res.json({
......@@ -411,7 +414,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 1 },
choices: [
{
message: { role: 'assistant', content: responseContent },
message: { role: 'assistant', content: formatResponseContent },
finish_reason: 'stop',
index: 0
}
......
projects/app/src/pages/chat/share.tsx
......@@ -87,6 +87,7 @@ const OutLink = (props: Props) => {
const setChatBoxData = useContextSelector(ChatItemContext, (v) => v.setChatBoxData);
const quoteData = useContextSelector(ChatItemContext, (v) => v.quoteData);
const setQuoteData = useContextSelector(ChatItemContext, (v) => v.setQuoteData);
const isResponseDetail = useContextSelector(ChatItemContext, (v) => v.isResponseDetail);
const chatRecords = useContextSelector(ChatRecordContext, (v) => v.chatRecords);
const totalRecordsCount = useContextSelector(ChatRecordContext, (v) => v.totalRecordsCount);
......@@ -162,7 +163,8 @@ const OutLink = (props: Props) => {
},
responseChatItemId,
chatId: completionChatId,
...outLinkAuthData
...outLinkAuthData,
retainDatasetCite: isResponseDetail
},
onMessage: generatingMessage,
abortCtrl: controller
......@@ -200,6 +202,7 @@ const OutLink = (props: Props) => {
chatId,
customVariables,
outLinkAuthData,
isResponseDetail,
onUpdateHistoryTitle,
setChatBoxData,
forbidLoadChat,
......
projects/app/src/service/events/generateQA.ts
......@@ -17,7 +17,7 @@ import {
} from '@fastgpt/service/common/string/tiktoken/index';
import { pushDataListToTrainingQueueByCollectionId } from '@fastgpt/service/core/dataset/training/controller';
import { loadRequestMessages } from '@fastgpt/service/core/chat/utils';
import { llmCompletionsBodyFormat, llmResponseToAnswerText } from '@fastgpt/service/core/ai/utils';
import { llmCompletionsBodyFormat, formatLLMResponse } from '@fastgpt/service/core/ai/utils';
import { LLMModelItemType } from '@fastgpt/global/core/ai/model.d';
import {
chunkAutoChunkSize,
......@@ -140,7 +140,7 @@ ${replaceVariable(Prompt_AgentQA.fixedText, { text })}`;
modelData
)
});
const { text: answer, usage } = await llmResponseToAnswerText(chatResponse);
const { text: answer, usage } = await formatLLMResponse(chatResponse);
const inputTokens = usage?.prompt_tokens || (await countGptMessagesTokens(messages));
const outputTokens = usage?.completion_tokens || (await countPromptTokens(answer));
......
projects/app/src/service/support/mcp/utils.ts
......@@ -37,6 +37,7 @@ import { saveChat } from '@fastgpt/service/core/chat/saveChat';
import { DispatchNodeResponseKeyEnum } from '@fastgpt/global/core/workflow/runtime/constants';
import { createChatUsage } from '@fastgpt/service/support/wallet/usage/controller';
import { UsageSourceEnum } from '@fastgpt/global/support/wallet/usage/constants';
import { removeDatasetCiteText } from '@fastgpt/service/core/ai/utils';
export const pluginNodes2InputSchema = (
nodes: { flowNodeType: FlowNodeTypeEnum; inputs: FlowNodeInputItemType[] }[]
......@@ -288,7 +289,7 @@ export const callMcpServerTool = async ({ key, toolName, inputs }: toolCallProps
})();
// Format response content
responseContent = responseContent.trim().replace(/\[\w+\]\(QUOTE\)/g, '');
responseContent = removeDatasetCiteText(responseContent.trim(), false);
return responseContent;
};
......
projects/app/src/web/common/api/fetch.ts
......@@ -132,7 +132,7 @@ export const streamFetch = ({
variables,
detail: true,
stream: true,
parseQuote: true
retainDatasetCite: data.retainDatasetCite ?? true
})
};
......
projects/app/src/web/support/user/login/constants.ts
......@@ -4,22 +4,3 @@ export enum LoginPageTypeEnum {
forgetPassword = 'forgetPassword',
wechat = 'wechat'
}
export const checkPasswordRule = (password: string) => {
const patterns = [
/\d/, // Contains digits
/[a-z]/, // Contains lowercase letters
/[A-Z]/, // Contains uppercase letters
/[!@#$%^&*()_+=-]/ // Contains special characters
];
const validChars = /^[\dA-Za-z!@#$%^&*()_+=-]{6,100}$/;
// Check length and valid characters
if (!validChars.test(password)) return false;
// Count how many patterns are satisfied
const matchCount = patterns.filter((pattern) => pattern.test(password)).length;
// Must satisfy at least 2 patterns
return matchCount >= 2;
};
test/cases/components/Markdown/utils.test.ts
......@@ -16,7 +16,7 @@ describe('Markdown utils', () => {
it('should convert quote references to proper markdown links', () => {
const input = '[123456789012345678901234]';
const expected = '[123456789012345678901234](QUOTE)';
const expected = '[123456789012345678901234](CITE)';
expect(mdTextFormat(input)).toBe(expected);
});
......@@ -35,7 +35,7 @@ describe('Markdown utils', () => {
const input =
'Math \\[x^2\\] with link https://test.com，and quote [123456789012345678901234]';
const expected =
'Math $$x^2$$ with link https://test.com ，and quote [123456789012345678901234](QUOTE)';
'Math $$x^2$$ with link https://test.com ，and quote [123456789012345678901234](CITE)';
expect(mdTextFormat(input)).toBe(expected);
});
});
......
test/cases/function/packages/global/common/string/chunks.json deleted 100644 → 0
[
"测试的呀,第一个表格\n\n| 序号 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 1 | 张三 | 25 | 工程师 | 北京 |\n| 2 | 李四 | 30 | 教师 | 上海 |\n| 3 | 王五 | 28 | 医生 | 广州 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 4 | 赵六 | 35 | 律师 | 深圳 |\n| 5 | 孙七 | 27 | 设计师 | 杭州 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 7 | 吴九 | 29 | 销售 | 武汉 |\n| 8 | 郑十 | 31 | 记者 | 南京 |\n| 9 | 刘一 | 33 | 建筑师 | 天津 |\n| 10 | 陈二 | 26 | 程序员 | 重庆 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1001 | 杨一 | 34 | 程序员 | 厦门 |\n| 1002 | 杨二 | 34 | 程序员 | 厦门 |\n| 1003 | 杨三 | 34 | 程序员 | 厦门 |",
"| 序号 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 1004 | 杨四 | 34 | 程序员 | 厦门 |\n| 1005 | 杨五 | 34 | 程序员 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 9 | 刘一 | 33 | 建筑师 | 天津 |\n| 10 | 陈二 | 26 | 程序员 | 重庆 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1001 | 杨一 | 34 | 程序员 | 厦门 |\n| 1002 | 杨二 | 34 | 程序员 | 厦门 |\n| 1003 | 杨三 | 34 | 程序员 | 厦门 |\n| 1004 | 杨四 | 34 | 程序员 | 厦门 |\n| 1005 | 杨五 | 34 | 程序员 | 厦门 |\n\n| 序号 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |\n| 1000 | 黄末 | 28 | 作家 | 厦门 |",
"这是第二段了，第二表格\n\n| 序号 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 1 | 张三 | 25 | 工程师 | 北京 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 2 | 李四 | 30 | 教师 | 上海 |\n| 3 | 王五 | 28 | 医生 | 广州 |\n| 4 | 赵六 | 35 | 律师 | 深圳 |\n| 5 | 孙七 | 27 | 设计师 | 杭州 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 7 | 吴九 | 29 | 销售 | 武汉 |\n| 8 | 郑十 | 31 | 记者 | 南京 |\n| 9 | 刘一 | 33 | 建筑师 | 天津 |\n| 10 | 陈二 | 26 | 程序员 | 重庆 |\n| 10004 | 黄末 | 28 | 作家 | 厦门 |\n| 10013 | 杨一 | 34 | 程序员 | 厦门 |\n\n\n结束了\n\n| 序号22 | 姓名 | 年龄 | 职业 | 城市 |\n| --- | --- | --- | --- | --- |\n| 1 | 张三 | 25 | 工程师 | 北京 |\n| 2 | 李四 | 30 | 教师 | 上海 |\n| 3 | 王五 | 28 | 医生 | 广州 |\n| 4 | 赵六 | 35 | 律师 | 深圳 |\n| 5 | 孙七 | 27 | 设计师 | 杭州 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 6 | 周八 | 32 | 会计 | 成都 |\n| 7 | 吴九 | 29 | 销售 | 武汉 |\n| 8 | 郑十 | 31 | 记者 | 南京 |\n| 9 | 刘一 | 33 | 建筑师 | 天津 |\n| 10 | 陈二 | 26 | 程序员 | 重庆 |\n| 10002 | 黄末 | 28 | 作家 | 厦门 |\n| 10012 | 杨一 | 34 | 程序员 | 厦门 |"
]
\ No newline at end of file
test/cases/global/common/string/utils.test.ts → test/cases/function/packages/global/common/string/password.test.ts
import { describe, expect, it } from 'vitest';
import { checkPasswordRule } from '@/web/support/user/login/constants';
import { checkPasswordRule } from '@fastgpt/global/common/string/password';
describe('PasswordRule', () => {
it('should be a valid password', () => {
......
test/cases/function/packages/global/common/string/textSplitter.test.ts
import { it, expect } from 'vitest'; // 必须显式导入
import { splitText2Chunks } from '@fastgpt/global/common/string/textSplitter';
import * as fs from 'fs';
const simpleChunks = (chunks: string[]) => {
return chunks.map((chunk) => chunk.replace(/\s+/g, ''));
......
packages/service/core/ai/utils.test.ts → test/cases/function/packages/service/core/ai/parseStreamResponse.test.ts
import { parseReasoningStreamContent } from './utils';
import { expect, test } from 'vitest';
import { CompletionFinishReason } from '@fastgpt/global/core/ai/type';
import { parseLLMStreamResponse } from '@fastgpt/service/core/ai/utils';
import { describe, expect, it } from 'vitest';
test('Parse reasoning stream content test', async () => {
describe('Parse reasoning stream content test', async () => {
const partList = [
{
data: [{ content: '你好1' }, { content: '你好2' }, { content: '你好3' }],
......@@ -117,29 +118,223 @@ test('Parse reasoning stream content test', async () => {
}
];
partList.forEach((part) => {
const { parsePart } = parseReasoningStreamContent();
// Remove think
partList.forEach((part, index) => {
it(`Reasoning test:${index}`, () => {
const { parsePart } = parseLLMStreamResponse();
let answer = '';
let reasoning = '';
part.data.forEach((item) => {
const formatPart = {
choices: [
{
delta: {
role: 'assistant',
content: item.content,
reasoning_content: item.reasoning_content
let answer = '';
let reasoning = '';
part.data.forEach((item) => {
const formatPart = {
choices: [
{
delta: {
role: 'assistant',
content: item.content,
reasoning_content: item.reasoning_content
}
}
}
]
};
const [reasoningContent, content] = parsePart(formatPart, true);
answer += content;
reasoning += reasoningContent;
]
};
const { reasoningContent, content } = parsePart({
part: formatPart,
parseThinkTag: true,
retainDatasetCite: false
});
answer += content;
reasoning += reasoningContent;
});
expect(answer).toBe(part.correct.answer);
expect(reasoning).toBe(part.correct.reasoning);
});
});
});
expect(answer).toBe(part.correct.answer);
expect(reasoning).toBe(part.correct.reasoning);
describe('Parse dataset cite content test', async () => {
const partList = [
{
// 完整的
data: [
{ content: '知识库' },
{ content: '问答系统' },
{ content: '[67e517e747' },
{ content: '67063e882d' },
{ content: '6861](CITE)' }
],
correct: {
content: '知识库问答系统[67e517e74767063e882d6861](CITE)',
responseContent: '知识库问答系统'
}
},
{
// 缺失结尾
data: [
{ content: '知识库问答系统' },
{ content: '[67e517e747' },
{ content: '67063e882d' },
{ content: '6861](CITE' }
],
correct: {
content: '知识库问答系统[67e517e74767063e882d6861](CITE',
responseContent: '知识库问答系统[67e517e74767063e882d6861](CITE'
}
},
{
// ObjectId 不正确
data: [
{ content: '知识库问答系统' },
{ content: '[67e517e747' },
{ content: '67882d' },
{ content: '6861](CITE)' }
],
correct: {
content: '知识库问答系统[67e517e74767882d6861](CITE)',
responseContent: '知识库问答系统[67e517e74767882d6861](CITE)'
}
},
{
// 其他链接
data: [{ content: '知识库' }, { content: '问答系统' }, { content: '[](https://fastgpt.cn)' }],
correct: {
content: '知识库问答系统[](https://fastgpt.cn)',
responseContent: '知识库问答系统[](https://fastgpt.cn)'
}
},
{
// 不完整的其他链接
data: [{ content: '知识库' }, { content: '问答系统' }, { content: '[](https://fastgp' }],
correct: {
content: '知识库问答系统[](https://fastgp',
responseContent: '知识库问答系统[](https://fastgp'
}
},
{
// 开头
data: [{ content: '[知识库' }, { content: '问答系统' }, { content: '[](https://fastgp' }],
correct: {
content: '[知识库问答系统[](https://fastgp',
responseContent: '[知识库问答系统[](https://fastgp'
}
},
{
// 结尾
data: [{ content: '知识库' }, { content: '问答系统' }, { content: '[' }],
correct: {
content: '知识库问答系统[',
responseContent: '知识库问答系统['
}
},
{
// 中间
data: [
{ content: '知识库' },
{ content: '问答系统' },
{ content: '[' },
{ content: '问答系统]' }
],
correct: {
content: '知识库问答系统[问答系统]',
responseContent: '知识库问答系统[问答系统]'
}
},
{
// 双链接
data: [
{ content: '知识库' },
{ content: '问答系统' },
{ content: '[](https://fastgpt.cn)' },
{ content: '[67e517e747' },
{ content: '67063e882d' },
{ content: '6861](CITE)' }
],
correct: {
content: '知识库问答系统[](https://fastgpt.cn)[67e517e74767063e882d6861](CITE)',
responseContent: '知识库问答系统[](https://fastgpt.cn)'
}
},
{
// 双链接缺失部分
data: [
{ content: '知识库' },
{ content: '问答系统' },
{ content: '[](https://fastgpt.cn)' },
{ content: '[67e517e747' },
{ content: '67063e882d' },
{ content: '6861](CIT' }
],
correct: {
content: '知识库问答系统[](https://fastgpt.cn)[67e517e74767063e882d6861](CIT',
responseContent: '知识库问答系统[](https://fastgpt.cn)[67e517e74767063e882d6861](CIT'
}
},
{
// 双Cite
data: [
{ content: '知识库' },
{ content: '问答系统' },
{ content: '[67e517e747' },
{ content: '67063e882d' },
{ content: '6861](CITE)' },
{ content: '[67e517e747' },
{ content: '67063e882d' },
{ content: '6861](CITE)' }
],
correct: {
content: '知识库问答系统[67e517e74767063e882d6861](CITE)[67e517e74767063e882d6861](CITE)',
responseContent: '知识库问答系统'
}
},
{
// 双Cite-第一个假Cite
data: [
{ content: '知识库' },
{ content: '问答系统' },
{ content: '[67e517e747' },
{ content: '6861](CITE)' },
{ content: '[67e517e747' },
{ content: '67063e882d' },
{ content: '6861](CITE)' }
],
correct: {
content: '知识库问答系统[67e517e7476861](CITE)[67e517e74767063e882d6861](CITE)',
responseContent: '知识库问答系统[67e517e7476861](CITE)'
}
}
];
partList.forEach((part, index) => {
it(`Dataset cite test: ${index}`, () => {
const { parsePart } = parseLLMStreamResponse();
let answer = '';
let responseContent = '';
part.data.forEach((item, index) => {
const formatPart = {
choices: [
{
delta: {
role: 'assistant',
content: item.content,
reasoning_content: ''
},
finish_reason: (index === part.data.length - 1
? 'stop'
: null) as CompletionFinishReason
}
]
};
const { content, responseContent: newResponseContent } = parsePart({
part: formatPart,
parseThinkTag: false,
retainDatasetCite: false
});
answer += content;
responseContent += newResponseContent;
});
expect(answer).toEqual(part.correct.content);
expect(responseContent).toEqual(part.correct.responseContent);
});
});
});
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or sign in to comment