perf: init model (#4610)

* fix: model config undefined value

* perf: init model

docSite/content/zh-cn/docs/development/upgrading/497.md

@@ -20,5 +20,5 @@ weight: 793
 1. 文件上传分块大小限制，避免超出 MongoDB 限制。
 2. 使用记录仪表盘，无法获取指定成员的使用统计。
 3. 仪表盘接口，因未考虑时区问题，统计异常。
-4. LLM 模型测试接口，无法测试未启用的 LLM。
+4. LLM 模型测试接口，无法测试未启用的 LLM。同时修复，模型测试接口会把模型自定义请求地址去除问题。
 

packages/service/common/system/log.ts

@@ -4,6 +4,12 @@ import { LogLevelEnum } from './log/constant';
 import { connectionMongo } from '../mongo/index';
 import { getMongoLog } from './log/schema';
 
+export enum EventTypeEnum {
+  outLinkBot = '[Outlink bot]',
+  feishuBot = '[Feishu bot]',
+  wxOffiaccount = '[Offiaccount bot]'
+}
+
 const logMap = {
   [LogLevelEnum.debug]: {
     levelLog: chalk.green('[Debug]')

packages/service/core/ai/config.ts

@@ -45,11 +45,13 @@ export const getAxiosConfig = (props?: { userKey?: OpenaiAccountType }) => {
 };
 
 export const createChatCompletion = async ({
+  modelData,
   body,
   userKey,
   timeout,
   options
 }: {
+  modelData?: LLMModelItemType;
   body: ChatCompletionCreateParamsNonStreaming | ChatCompletionCreateParamsStreaming;
   userKey?: OpenaiAccountType;
   timeout?: number;
@@ -70,10 +72,11 @@ export const createChatCompletion = async ({
 > => {
   try {
     // Rewrite model
-    const modelConstantsData = getLLMModel(body.model);
+    const modelConstantsData = modelData || getLLMModel(body.model);
     if (!modelConstantsData) {
       return Promise.reject(`${body.model} not found`);
     }
+    body.model = modelConstantsData.model;
 
     const formatTimeout = timeout ? timeout : body.stream ? 60000 : 600000;
     const ai = getAIApi({

packages/service/core/ai/config/utils.ts

@@ -23,13 +23,7 @@ import {
 } from '../../../common/system/config/controller';
 import { delay } from '@fastgpt/global/common/system/utils';
 
-/* 
+const getModelConfigBaseUrl = () => {
-  TODO: 分优先级读取：
-  1. 有外部挂载目录，则读取外部的
-  2. 没有外部挂载目录，则读取本地的。然后试图拉取云端的进行覆盖。
-*/
-export const loadSystemModels = async (init = false) => {
-  const getProviderList = () => {
   const currentFileUrl = new URL(import.meta.url);
   const filePath = decodeURIComponent(
     process.platform === 'win32'
@@ -37,15 +31,21 @@ export const loadSystemModels = async (init = false) => {
       : currentFileUrl.pathname
   );
   const modelsPath = path.join(path.dirname(filePath), 'provider');
+  return modelsPath;
+};
 
-    return fs.readdirSync(modelsPath) as string[];
+/* 
-  };
+  TODO: 分优先级读取：
+  1. 有外部挂载目录，则读取外部的
+  2. 没有外部挂载目录，则读取本地的。然后试图拉取云端的进行覆盖。
+*/
+export const loadSystemModels = async (init = false) => {
   const pushModel = (model: SystemModelItemType) => {
     global.systemModelList.push(model);
 
     if (model.isActive) {
       global.systemActiveModelList.push(model);
-    }
+
       if (model.type === ModelTypeEnum.llm) {
         global.llmModelMap.set(model.model, model);
         global.llmModelMap.set(model.name, model);
@@ -83,6 +83,7 @@ export const loadSystemModels = async (init = false) => {
           global.systemDefaultModel.rerank = model;
         }
       }
+    }
   };
 
   if (!init && global.systemModelList) return;
@@ -99,9 +100,10 @@ export const loadSystemModels = async (init = false) => {
 
   try {
     const dbModels = await MongoSystemModel.find({}).lean();
-    const providerList = getProviderList();
 
-    // System model
+    // Load system model from local
+    const modelsPath = getModelConfigBaseUrl();
+    const providerList = fs.readdirSync(modelsPath) as string[];
     await Promise.all(
       providerList.map(async (name) => {
         const fileContent = (await import(`./provider/${name}`))?.default as {

packages/service/core/ai/model.ts

@@ -1,3 +1,4 @@
+import { cloneDeep } from 'lodash';
 import { SystemModelItemType } from './type';
 
 export const getDefaultLLMModel = () => global?.systemDefaultModel.llm!;
@@ -53,5 +54,5 @@ export const findAIModel = (model: string): SystemModelItemType | undefined => {
   );
 };
 export const findModelFromAlldata = (model: string) => {
-  return global.systemModelList.find((item) => item.model === model);
+  return cloneDeep(global.systemModelList.find((item) => item.model === model));
 };

packages/web/components/common/Input/NumberInput/index.tsx

@@ -26,9 +26,9 @@ const MyNumberInput = (props: Props) => {
     <NumberInput
       {...restProps}
       onBlur={(e) => {
-        const numE = Number(e.target.value);
+        const numE = e.target.value === '' ? '' : Number(e.target.value);
         if (onBlur) {
-          if (isNaN(numE)) {
+          if (numE === '') {
             // @ts-ignore
             onBlur('');
           } else {
@@ -46,9 +46,9 @@ const MyNumberInput = (props: Props) => {
         }
       }}
       onChange={(e) => {
-        const numE = Number(e);
+        const numE = e === '' ? '' : Number(e);
         if (onChange) {
-          if (isNaN(numE)) {
+          if (numE === '') {
             // @ts-ignore
             onChange('');
           } else {
@@ -62,6 +62,7 @@ const MyNumberInput = (props: Props) => {
               value: numE
             }
           };
+
           register(name).onChange(event);
         }
       }}

packages/web/i18n/en/account_model.json

@@ -30,7 +30,7 @@
   "log_status": "Status",
   "mapping": "Model Mapping",
   "mapping_tip": "A valid Json is required. \nThe model can be mapped when sending a request to the actual address. \nFor example:\n{\n  \n  \"gpt-4o\": \"gpt-4o-test\"\n\n}\n\nWhen FastGPT requests the gpt-4o model, the gpt-4o-test model is sent to the actual address, instead of gpt-4o.",
-  "maxToken_tip": "The model max_tokens parameter, if left blank, means that the model does not support it.",
+  "maxToken_tip": "Model max_tokens parameter",
   "max_temperature_tip": "If the model temperature parameter is not filled in, it means that the model does not support the temperature parameter.",
   "model": "Model",
   "model_name": "Model name",

packages/web/i18n/zh-CN/account_model.json

@@ -30,7 +30,7 @@
   "log_status": "状态",
   "mapping": "模型映射",
   "mapping_tip": "需填写一个有效 Json。可在向实际地址发送请求时，对模型进行映射。例如：\n{\n  \"gpt-4o\": \"gpt-4o-test\"\n}\n当 FastGPT 请求 gpt-4o 模型时，会向实际地址发送 gpt-4o-test 的模型，而不是 gpt-4o。",
-  "maxToken_tip": "模型 max_tokens 参数，如果留空，则代表模型不支持该参数。",
+  "maxToken_tip": "模型 max_tokens 参数",
   "max_temperature_tip": "模型 temperature 参数，不填则代表模型不支持 temperature 参数。",
   "model": "模型",
   "model_name": "模型名",

packages/web/i18n/zh-Hant/account_model.json

@@ -28,7 +28,7 @@
   "log_status": "狀態",
   "mapping": "模型對映",
   "mapping_tip": "需填寫一個有效 Json。\n可在向實際地址傳送請求時，對模型進行對映。\n例如：\n{\n  \n  \"gpt-4o\": \"gpt-4o-test\"\n\n}\n\n當 FastGPT 請求 gpt-4o 模型時，會向實際地址傳送 gpt-4o-test 的模型，而不是 gpt-4o。",
-  "maxToken_tip": "模型 max_tokens 參數，如果留空，則代表模型不支援該參數。",
+  "maxToken_tip": "模型 max_tokens 參數",
   "max_temperature_tip": "模型 temperature 參數，不填則代表模型不支援 temperature 參數。",
   "model": "模型",
   "model_name": "模型名",

projects/app/src/pageComponents/account/model/AddModelBox.tsx

@@ -356,7 +356,12 @@ export const ModelEditModal = ({
                       </Td>
                       <Td textAlign={'right'}>
                         <Flex justifyContent={'flex-end'}>
-                          <MyNumberInput register={register} name="maxResponse" {...InputStyles} />
+                          <MyNumberInput
+                            min={2000}
+                            register={register}
+                            name="maxResponse"
+                            {...InputStyles}
+                          />
                         </Flex>
                       </Td>
                     </Tr>
@@ -372,6 +377,7 @@ export const ModelEditModal = ({
                           <MyNumberInput
                             register={register}
                             name="maxTemperature"
+                            min={0}
                             step={0.1}
                             {...InputStyles}
                           />

projects/app/src/pages/api/core/ai/model/test.ts

@@ -79,6 +79,7 @@ const testLLMModel = async (model: LLMModelItemType, headers: Record<string, str
   );
 
   const { response, isStreamResponse } = await createChatCompletion({
+    modelData: model,
     body: requestBody,
     options: {
       headers: {