✨ feat: add Hunyuan(Tencent) model provider (lobehub#4147)

* ✨ feat: add Hunyuan(Tencent) model provider

* 💄 style: add `HUNYUAN_MODEL_LIST` support

* 💄 style: update model price & provider description

* 💄 style: enable `hunyuan-standard-256K` as default, sort model provider list

* 💄 style: update model tokens info

* 💄 style: update pricing

* 💄 style: update `hunyuan-turbo` model info

Dockerfile

@@ -135,6 +135,8 @@ ENV \
     GOOGLE_API_KEY="" GOOGLE_PROXY_URL="" \
     # Groq
     GROQ_API_KEY="" GROQ_MODEL_LIST="" GROQ_PROXY_URL="" \
+    # Hunyuan
+    HUNYUAN_API_KEY="" HUNYUAN_MODEL_LIST="" \
     # Minimax
     MINIMAX_API_KEY="" \
     # Mistral

Dockerfile.database

@@ -167,6 +167,8 @@ ENV \
     GOOGLE_API_KEY="" GOOGLE_PROXY_URL="" \
     # Groq
     GROQ_API_KEY="" GROQ_MODEL_LIST="" GROQ_PROXY_URL="" \
+    # Hunyuan
+    HUNYUAN_API_KEY="" HUNYUAN_MODEL_LIST="" \
     # Minimax
     MINIMAX_API_KEY="" \
     # Mistral

src/app/(main)/settings/llm/ProviderList/providers.tsx

@@ -9,6 +9,7 @@ import {
   FireworksAIProviderCard,
   GoogleProviderCard,
   GroqProviderCard,
+  HunyuanProviderCard,
   MinimaxProviderCard,
   MistralProviderCard,
   MoonshotProviderCard,
@@ -60,6 +61,7 @@ export const useProviderList = (): ProviderItem[] => {
       Ai21ProviderCard,
       UpstageProviderCard,
       QwenProviderCard,
+      HunyuanProviderCard,
       SparkProviderCard,
       ZhiPuProviderCard,
       ZeroOneProviderCard,

src/app/api/chat/agentRuntime.ts

@@ -244,6 +244,13 @@ const getLlmOptionsFromPayload = (provider: string, payload: JWTPayload) => {
 
       const apiKey = apiKeyManager.pick(payload?.apiKey || AI21_API_KEY);
 
+      return { apiKey };
+    }
+    case ModelProvider.Hunyuan: {
+      const { HUNYUAN_API_KEY } = getLLMConfig();
+
+      const apiKey = apiKeyManager.pick(payload?.apiKey || HUNYUAN_API_KEY);
+
       return { apiKey };
     }
   }

src/config/llm.ts

@@ -117,6 +117,10 @@ export const getLLMConfig = () => {
 
       ENABLED_AI21: z.boolean(),
       AI21_API_KEY: z.string().optional(),
+
+      ENABLED_HUNYUAN: z.boolean(),
+      HUNYUAN_API_KEY: z.string().optional(),
+      HUNYUAN_MODEL_LIST: z.string().optional(),
     },
     runtimeEnv: {
       API_KEY_SELECT_MODE: process.env.API_KEY_SELECT_MODE,
@@ -231,6 +235,10 @@ export const getLLMConfig = () => {
 
       ENABLED_AI21: !!process.env.AI21_API_KEY,
       AI21_API_KEY: process.env.AI21_API_KEY,
+
+      ENABLED_HUNYUAN: !!process.env.HUNYUAN_API_KEY,
+      HUNYUAN_API_KEY: process.env.HUNYUAN_API_KEY,
+      HUNYUAN_MODEL_LIST: process.env.HUNYUAN_MODEL_LIST,
     },
   });
 };

src/config/modelProviders/hunyuan.ts

@@ -0,0 +1,137 @@
+import { ModelProviderCard } from '@/types/llm';
+
+// ref https://cloud.tencent.com/document/product/1729/104753
+const Hunyuan: ModelProviderCard = {
+  chatModels: [
+    {
+      description: '升级为 MOE 结构，上下文窗口为 256k ，在 NLP，代码，数学，行业等多项评测集上领先众多开源模型。',
+      displayName: 'Hunyuan Lite',
+      enabled: true,
+      id: 'hunyuan-lite',
+      maxOutput: 6000,
+      pricing: {
+        currency: 'CNY',
+        input: 0,
+        output: 0,
+      },
+      tokens: 256_000,
+    },
+    {
+      description: '采用更优的路由策略，同时缓解了负载均衡和专家趋同的问题。长文方面，大海捞针指标达到99.9%。MOE-32K 性价比相对更高，在平衡效果、价格的同时，可对实现对长文本输入的处理。',
+      displayName: 'Hunyuan Standard',
+      enabled: true,
+      id: 'hunyuan-standard',
+      maxOutput: 2000,
+      pricing: {
+        currency: 'CNY',
+        input: 4.5,
+        output: 5,
+      },
+      tokens: 32_000,
+    },
+    {
+      description: '采用更优的路由策略，同时缓解了负载均衡和专家趋同的问题。长文方面，大海捞针指标达到99.9%。MOE-256K 在长度和效果上进一步突破，极大的扩展了可输入长度。',
+      displayName: 'Hunyuan Standard 256K',
+      enabled: true,
+      id: 'hunyuan-standard-256K',
+      maxOutput: 6000,
+      pricing: {
+        currency: 'CNY',
+        input: 15,
+        output: 60,
+      },
+      tokens: 256_000,
+    },
+    {
+      description: '混元全新一代大语言模型的预览版，采用全新的混合专家模型（MoE）结构，相比hunyuan-pro推理效率更快，效果表现更强。',
+      displayName: 'Hunyuan Turbo',
+      enabled: true,
+      functionCall: true,
+      id: 'hunyuan-turbo',
+      maxOutput: 4000,
+      pricing: {
+        currency: 'CNY',
+        input: 15,
+        output: 50,
+      },
+      tokens: 32_000,
+    },
+    {
+      description: '万亿级参数规模 MOE-32K 长文模型。在各种 benchmark 上达到绝对领先的水平，复杂指令和推理，具备复杂数学能力，支持 functioncall，在多语言翻译、金融法律医疗等领域应用重点优化。',
+      displayName: 'Hunyuan Pro',
+      enabled: true,
+      functionCall: true,
+      id: 'hunyuan-pro',
+      maxOutput: 4000,
+      pricing: {
+        currency: 'CNY',
+        input: 30,
+        output: 100,
+      },
+      tokens: 32_000,
+    },
+    {
+      description: '混元最新代码生成模型，经过 200B 高质量代码数据增训基座模型，迭代半年高质量 SFT 数据训练，上下文长窗口长度增大到 8K，五大语言代码生成自动评测指标上位居前列；五大语言10项考量各方面综合代码任务人工高质量评测上，性能处于第一梯队',
+      displayName: 'Hunyuan Code',
+      enabled: true,
+      id: 'hunyuan-code',
+      maxOutput: 4000,
+      pricing: {
+        currency: 'CNY',
+        input: 4,
+        output: 8,
+      },
+      tokens: 8000,
+    },
+    {
+      description: '混元最新多模态模型，支持图片+文本输入生成文本内容。',
+      displayName: 'Hunyuan Vision',
+      enabled: true,
+      id: 'hunyuan-vision',
+      maxOutput: 4000,
+      pricing: {
+        currency: 'CNY',
+        input: 18,
+        output: 18,
+      },
+      tokens: 8000,
+      vision: true,
+    },
+    {
+      description: '混元最新 MOE 架构 FunctionCall 模型，经过高质量的 FunctionCall 数据训练，上下文窗口达 32K，在多个维度的评测指标上处于领先。',
+      displayName: 'Hunyuan FunctionCall',
+      functionCall: true,
+      id: 'hunyuan-functioncall',
+      maxOutput: 4000,
+      pricing: {
+        currency: 'CNY',
+        input: 4,
+        output: 8,
+      },
+      tokens: 32_000,
+    },
+    {
+      description: '混元最新版角色扮演模型，混元官方精调训练推出的角色扮演模型，基于混元模型结合角色扮演场景数据集进行增训，在角色扮演场景具有更好的基础效果。',
+      displayName: 'Hunyuan Role',
+      id: 'hunyuan-role',
+      maxOutput: 4000,
+      pricing: {
+        currency: 'CNY',
+        input: 4,
+        output: 8,
+      },
+      tokens: 8000,
+    },
+  ],
+  checkModel: 'hunyuan-lite',
+  description:
+    '由腾讯研发的大语言模型，具备强大的中文创作能力，复杂语境下的逻辑推理能力，以及可靠的任务执行能力',
+  disableBrowserRequest: true,
+  id: 'hunyuan',
+  modelList: { showModelFetcher: true },
+  modelsUrl: 'https://cloud.tencent.com/document/product/1729/104753',
+  name: 'Hunyuan',
+  url: 'https://hunyuan.tencent.com',
+};
+
+export default Hunyuan;

src/config/modelProviders/index.ts

@@ -11,6 +11,7 @@ import FireworksAIProvider from './fireworksai';
 import GithubProvider from './github';
 import GoogleProvider from './google';
 import GroqProvider from './groq';
+import HunyuanProvider from './hunyuan';
 import MinimaxProvider from './minimax';
 import MistralProvider from './mistral';
 import MoonshotProvider from './moonshot';
@@ -57,6 +58,7 @@ export const LOBE_DEFAULT_MODEL_LIST: ChatModelCard[] = [
   UpstageProvider.chatModels,
   SparkProvider.chatModels,
   Ai21Provider.chatModels,
+  HunyuanProvider.chatModels,
 ].flat();
 
 export const DEFAULT_MODEL_PROVIDER_LIST = [
@@ -78,6 +80,7 @@ export const DEFAULT_MODEL_PROVIDER_LIST = [
   Ai21Provider,
   UpstageProvider,
   QwenProvider,
+  HunyuanProvider,
   SparkProvider,
   ZhiPuProvider,
   ZeroOneProvider,
@@ -110,6 +113,7 @@ export { default as FireworksAIProviderCard } from './fireworksai';
 export { default as GithubProviderCard } from './github';
 export { default as GoogleProviderCard } from './google';
 export { default as GroqProviderCard } from './groq';
+export { default as HunyuanProviderCard } from './hunyuan';
 export { default as MinimaxProviderCard } from './minimax';
 export { default as MistralProviderCard } from './mistral';
 export { default as MoonshotProviderCard } from './moonshot';

src/const/settings/llm.ts

@@ -9,6 +9,7 @@ import {
   GithubProviderCard,
   GoogleProviderCard,
   GroqProviderCard,
+  HunyuanProviderCard,
   MinimaxProviderCard,
   MistralProviderCard,
   MoonshotProviderCard,
@@ -75,6 +76,10 @@ export const DEFAULT_LLM_CONFIG: UserModelProviderConfig = {
     enabled: false,
     enabledModels: filterEnabledModels(GroqProviderCard),
   },
+  hunyuan: {
+    enabled: false,
+    enabledModels: filterEnabledModels(HunyuanProviderCard),
+  },
   minimax: {
     enabled: false,
     enabledModels: filterEnabledModels(MinimaxProviderCard),

src/libs/agent-runtime/AgentRuntime.ts

@@ -14,6 +14,7 @@ import { LobeFireworksAI } from './fireworksai';
 import { LobeGithubAI } from './github';
 import { LobeGoogleAI } from './google';
 import { LobeGroq } from './groq';
+import { LobeHunyuanAI } from './hunyuan';
 import { LobeMinimaxAI } from './minimax';
 import { LobeMistralAI } from './mistral';
 import { LobeMoonshotAI } from './moonshot';
@@ -133,6 +134,7 @@ class AgentRuntime {
       github: Partial<ClientOptions>;
       google: { apiKey?: string; baseURL?: string };
       groq: Partial<ClientOptions>;
+      hunyuan: Partial<ClientOptions>;
       minimax: Partial<ClientOptions>;
       mistral: Partial<ClientOptions>;
       moonshot: Partial<ClientOptions>;
@@ -300,6 +302,11 @@ class AgentRuntime {
         runtimeModel = new LobeAi21AI(params.ai21);
         break;
       }
+
+      case ModelProvider.Hunyuan: {
+        runtimeModel = new LobeHunyuanAI(params.hunyuan);
+        break;
+      }
     }
 
     return new AgentRuntime(runtimeModel);