V4.8.17 feature (#3493)

* split tokens into input and output (#3477)

* split tokens into input and output

* query extension & tool call & question guide

* fix

* perf: input and output tokens

* perf: tool call if else

* perf: remove code

* fix: extract usage count

* fix: qa usage count

---------

Co-authored-by: heheer <[email protected]>

Author: c121914yu

docSite/content/zh-cn/docs/development/upgrading/4817.md

@@ -28,6 +28,11 @@ curl --location --request POST 'https://{{host}}/api/admin/initv4817' \
 
 会将用户绑定的 OpenAI 账号移动到团队中。
 
+
+## 调整 completions 接口返回值
+
+/api/v1/chat/completions 接口返回值调整，对话节点、工具节点等使用到模型的节点，将不再返回 `tokens` 字段，改为返回 `inputTokens` 和 `outputTokens` 字段，分别表示输入和输出的 Token 数量。
+
 ## 完整更新内容
 
 1. 新增 - 简易模式工具调用支持数组类型插件。
@@ -36,10 +41,12 @@ curl --location --request POST 'https://{{host}}/api/admin/initv4817' \
 4. 新增 - 商业版支持后台配置模板市场。
 5. 新增 - 商业版支持后台配置自定义工作流变量，用于与业务系统鉴权打通。
 6. 新增 - 搜索测试接口支持问题优化。
-7. 优化 - Markdown 大小测试，超出 20 万字符不使用 Markdown 组件，避免崩溃。
-8. 优化 - 知识库搜索参数，滑动条支持输入模式，可以更精准的控制。
-9. 优化 - 可用模型展示
-10. 优化 - Mongo 查询语句，增加 virtual 字段。
-11. 修复 - 文件返回接口缺少 Content-Length 头，导致通过非同源文件上传时，阿里 vision 模型无法识别图片。
-12. 修复 - 去除判断器两端字符串隐藏换行符，避免判断器失效。
-13. 修复 - 变量更新节点，手动输入更新内容时候，非字符串类型数据类型无法自动转化。
+7. 新增 - 工作流中 Input Token 和 Output Token 分开记录展示。并修复部分请求未记录输出 Token 计费问题。
+8. 优化 - Markdown 大小测试，超出 20 万字符不使用 Markdown 组件，避免崩溃。
+9. 优化 - 知识库搜索参数，滑动条支持输入模式，可以更精准的控制。
+10. 优化 - 可用模型展示UI。
+11. 优化 - Mongo 查询语句，增加 virtual 字段。
+12. 修复 - 文件返回接口缺少 Content-Length 头，导致通过非同源文件上传时，阿里 vision 模型无法识别图片。
+13. 修复 - 去除判断器两端字符串隐藏换行符，避免判断器失效。
+14. 修复 - 变量更新节点，手动输入更新内容时候，非字符串类型数据类型无法自动转化。
+15. 修复 - 豆包模型无法工具调用。

packages/global/core/ai/model.d.ts

@@ -1,6 +1,13 @@
 import type { ModelProviderIdType } from './provider';
 
-export type LLMModelItemType = {
+type PriceType = {
+  charsPointsPrice?: number; // 1k chars=n points; 60s=n points;
+
+  // If inputPrice is set, the input-output charging scheme is adopted
+  inputPrice?: number; // 1k tokens=n points
+  outputPrice?: number; // 1k tokens=n points
+};
+export type LLMModelItemType = PriceType & {
   provider: ModelProviderIdType;
   model: string;
   name: string;
@@ -10,8 +17,6 @@ export type LLMModelItemType = {
   quoteMaxToken: number;
   maxTemperature: number;
 
-  charsPointsPrice: number; // 1k chars=n points
-
   censor?: boolean;
   vision?: boolean;
 
@@ -33,13 +38,12 @@ export type LLMModelItemType = {
   fieldMap?: Record<string, string>;
 };
 
-export type VectorModelItemType = {
+export type VectorModelItemType = PriceType & {
   provider: ModelProviderIdType;
   model: string; // model name
   name: string; // show name
   avatar?: string;
   defaultToken: number; // split text default token
-  charsPointsPrice: number; // 1k tokens=n points
   maxToken: number; // model max token
   weight: number; // training weight
   hidden?: boolean; // Disallow creation
@@ -48,25 +52,22 @@ export type VectorModelItemType = {
   queryConfig?: Record<string, any>; // Custom parameters for query
 };
 
-export type ReRankModelItemType = {
+export type ReRankModelItemType = PriceType & {
   model: string;
   name: string;
-  charsPointsPrice: number;
   requestUrl: string;
   requestAuth: string;
 };
 
-export type AudioSpeechModelType = {
+export type AudioSpeechModelType = PriceType & {
   provider: ModelProviderIdType;
   model: string;
   name: string;
-  charsPointsPrice: number;
   voices: { label: string; value: string; bufferId: string }[];
 };
 
-export type STTModelType = {
+export type STTModelType = PriceType & {
   provider: ModelProviderIdType;
   model: string;
   name: string;
-  charsPointsPrice: number; // 60s = n points
 };

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

@@ -107,7 +107,9 @@ export type DispatchNodeResponseType = {
   mergeSignId?: string;
 
   // bill
-  tokens?: number;
+  tokens?: number; // deprecated
+  inputTokens?: number;
+  outputTokens?: number;
   model?: string;
   contextTotalLen?: number;
   totalPoints?: number;
@@ -157,6 +159,8 @@ export type DispatchNodeResponseType = {
 
   // tool
   toolCallTokens?: number;
+  toolCallInputTokens?: number;
+  toolCallOutputTokens?: number;
   toolDetail?: ChatHistoryItemResType[];
   toolStop?: boolean;
 

packages/global/support/wallet/bill/type.d.ts

@@ -23,7 +23,8 @@ export type BillSchemaType = {
 };
 
 export type ChatNodeUsageType = {
-  tokens?: number;
+  inputTokens?: number;
+  outputTokens?: number;
   totalPoints: number;
   moduleName: string;
   model?: string;

packages/global/support/wallet/usage/type.d.ts

@@ -2,9 +2,13 @@ import { CreateUsageProps } from './api';
 import { UsageSourceEnum } from './constants';
 
 export type UsageListItemCountType = {
-  tokens?: number;
+  inputTokens?: number;
+  outputTokens?: number;
   charsLength?: number;
   duration?: number;
+
+  // deprecated
+  tokens?: number;
 };
 export type UsageListItemType = UsageListItemCountType & {
   moduleName: string;

packages/plugins/src/feishu/template.json

@@ -2,7 +2,7 @@
   "author": "",
   "version": "488",
   "name": "飞书 webhook",
-  "avatar": "/appMarketTemplates/plugin-feishu/avatar.svg",
+  "avatar": "core/app/templates/plugin-feishu",
   "intro": "向飞书机器人发起 webhook 请求。",
   "courseUrl": "https://open.feishu.cn/document/client-docs/bot-v3/add-custom-bot#f62e72d5",
   "showStatus": false,

packages/service/common/system/tools.ts

@@ -15,6 +15,9 @@ export const initFastGPTConfig = (config?: FastGPTConfigFileType) => {
   global.subPlans = config.subPlans;
 
   global.llmModels = config.llmModels;
+  global.llmModelPriceType = global.llmModels.some((item) => typeof item.inputPrice === 'number')
+    ? 'IO'
+    : 'Tokens';
   global.vectorModels = config.vectorModels;
   global.audioSpeechModels = config.audioSpeechModels;
   global.whisperModel = config.whisperModel;

packages/service/core/ai/functions/createQuestionGuide.ts

@@ -1,6 +1,6 @@
 import type { ChatCompletionMessageParam } from '@fastgpt/global/core/ai/type.d';
 import { createChatCompletion } from '../config';
-import { countGptMessagesTokens } from '../../../common/string/tiktoken/index';
+import { countGptMessagesTokens, countPromptTokens } from '../../../common/string/tiktoken/index';
 import { loadRequestMessages } from '../../chat/utils';
 import { llmCompletionsBodyFormat } from '../utils';
 import {
@@ -20,7 +20,8 @@ export async function createQuestionGuide({
   customPrompt?: string;
 }): Promise<{
   result: string[];
-  tokens: number;
+  inputTokens: number;
+  outputTokens: number;
 }> {
   const concatMessages: ChatCompletionMessageParam[] = [
     ...messages,
@@ -29,17 +30,18 @@ export async function createQuestionGuide({
       content: `${customPrompt || PROMPT_QUESTION_GUIDE}\n${PROMPT_QUESTION_GUIDE_FOOTER}`
     }
   ];
+  const requestMessages = await loadRequestMessages({
+    messages: concatMessages,
+    useVision: false
+  });
 
   const { response: data } = await createChatCompletion({
     body: llmCompletionsBodyFormat(
       {
         model,
         temperature: 0.1,
         max_tokens: 200,
-        messages: await loadRequestMessages({
-          messages: concatMessages,
-          useVision: false
-        }),
+        messages: requestMessages,
         stream: false
       },
       model
@@ -51,13 +53,15 @@ export async function createQuestionGuide({
   const start = answer.indexOf('[');
   const end = answer.lastIndexOf(']');
 
-  const tokens = await countGptMessagesTokens(concatMessages);
+  const inputTokens = await countGptMessagesTokens(requestMessages);
+  const outputTokens = await countPromptTokens(answer);
 
   if (start === -1 || end === -1) {
     addLog.warn('Create question guide error', { answer });
     return {
       result: [],
-      tokens: 0
+      inputTokens: 0,
+      outputTokens: 0
     };
   }
 
@@ -69,14 +73,16 @@ export async function createQuestionGuide({
   try {
     return {
       result: json5.parse(jsonStr),
-      tokens
+      inputTokens,
+      outputTokens
     };
   } catch (error) {
     console.log(error);
 
     return {
       result: [],
-      tokens: 0
+      inputTokens: 0,
+      outputTokens: 0
     };
   }
 }

packages/service/core/ai/functions/queryExtension.ts

@@ -1,7 +1,7 @@
 import { replaceVariable } from '@fastgpt/global/common/string/tools';
 import { createChatCompletion } from '../config';
 import { ChatItemType } from '@fastgpt/global/core/chat/type';
-import { countGptMessagesTokens } from '../../../common/string/tiktoken/index';
+import { countGptMessagesTokens, countPromptTokens } from '../../../common/string/tiktoken/index';
 import { chatValue2RuntimePrompt } from '@fastgpt/global/core/chat/adapt';
 import { getLLMModel } from '../model';
 import { llmCompletionsBodyFormat } from '../utils';
@@ -121,7 +121,8 @@ export const queryExtension = async ({
   rawQuery: string;
   extensionQueries: string[];
   model: string;
-  tokens: number;
+  inputTokens: number;
+  outputTokens: number;
 }> => {
   const systemFewShot = chatBg
     ? `Q: 对话背景。
@@ -166,7 +167,8 @@ A: ${chatBg}
       rawQuery: query,
       extensionQueries: [],
       model,
-      tokens: 0
+      inputTokens: 0,
+      outputTokens: 0
     };
   }
 
@@ -181,15 +183,17 @@ A: ${chatBg}
       rawQuery: query,
       extensionQueries: Array.isArray(queries) ? queries : [],
       model,
-      tokens: await countGptMessagesTokens(messages)
+      inputTokens: await countGptMessagesTokens(messages),
+      outputTokens: await countPromptTokens(answer)
     };
   } catch (error) {
     addLog.error(`Query extension error`, error);
     return {
       rawQuery: query,
       extensionQueries: [],
       model,
-      tokens: 0
+      inputTokens: 0,
+      outputTokens: 0
     };
   }
 };

packages/service/core/ai/model.ts

@@ -4,6 +4,7 @@ export const getLLMModel = (model?: string) => {
     global.llmModels[0]
   );
 };
+
 export const getDatasetModel = (model?: string) => {
   return (
     global.llmModels