✨ feat: add SambaNova provider support (lobehub#6218)

* ✨ feat: add SambaNova provider support

* 🐛 fix: fix build error

* 🔨 chore: disable Browser Request & `DeepSeek-R1` model

Dockerfile

@@ -205,6 +205,8 @@ ENV \
     PERPLEXITY_API_KEY="" PERPLEXITY_MODEL_LIST="" PERPLEXITY_PROXY_URL="" \
     # Qwen
     QWEN_API_KEY="" QWEN_MODEL_LIST="" QWEN_PROXY_URL="" \
+    # SambaNova
+    SAMBANOVA_API_KEY="" SAMBANOVA_MODEL_LIST="" \
     # SenseNova
     SENSENOVA_API_KEY="" SENSENOVA_MODEL_LIST="" \
     # SiliconCloud

Dockerfile.database

@@ -242,6 +242,8 @@ ENV \
     PERPLEXITY_API_KEY="" PERPLEXITY_MODEL_LIST="" PERPLEXITY_PROXY_URL="" \
     # Qwen
     QWEN_API_KEY="" QWEN_MODEL_LIST="" QWEN_PROXY_URL="" \
+    # SambaNova
+    SAMBANOVA_API_KEY="" SAMBANOVA_MODEL_LIST="" \
     # SenseNova
     SENSENOVA_API_KEY="" SENSENOVA_MODEL_LIST="" \
     # SiliconCloud

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

@@ -22,6 +22,7 @@ import {
   OpenRouterProviderCard,
   PerplexityProviderCard,
   QwenProviderCard,
+  SambaNovaProviderCard,
   SenseNovaProviderCard,
   SiliconCloudProviderCard,
   SparkProviderCard,
@@ -79,6 +80,7 @@ export const useProviderList = (): ProviderItem[] => {
       UpstageProviderCard,
       XAIProviderCard,
       JinaProviderCard,
+      SambaNovaProviderCard,
       QwenProviderCard,
       WenxinProviderCard,
       HunyuanProviderCard,

src/config/aiModels/index.ts

@@ -31,6 +31,7 @@ import { default as openai } from './openai';
 import { default as openrouter } from './openrouter';
 import { default as perplexity } from './perplexity';
 import { default as qwen } from './qwen';
+import { default as sambanova } from './sambanova';
 import { default as sensenova } from './sensenova';
 import { default as siliconcloud } from './siliconcloud';
 import { default as spark } from './spark';
@@ -97,6 +98,7 @@ export const LOBE_DEFAULT_MODEL_LIST = buildDefaultModelList({
   openrouter,
   perplexity,
   qwen,
+  sambanova,
   sensenova,
   siliconcloud,
   spark,
@@ -144,6 +146,7 @@ export { default as openai } from './openai';
 export { default as openrouter } from './openrouter';
 export { default as perplexity } from './perplexity';
 export { default as qwen } from './qwen';
+export { default as sambanova } from './sambanova';
 export { default as sensenova } from './sensenova';
 export { default as siliconcloud } from './siliconcloud';
 export { default as spark } from './spark';

src/config/aiModels/sambanova.ts

@@ -0,0 +1,195 @@
+import { AIChatModelCard } from '@/types/aiModel';
+
+const sambanovaChatModels: AIChatModelCard[] = [
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 16_000,
+    description: 'Llama 3.3 是 Llama 系列最先进的多语言开源大型语言模型，以极低成本体验媲美 405B 模型的性能。基于 Transformer 结构，并通过监督微调（SFT）和人类反馈强化学习（RLHF）提升有用性和安全性。其指令调优版本专为多语言对话优化，在多项行业基准上表现优于众多开源和封闭聊天模型。知识截止日期为 2023 年 12 月',
+    displayName: 'Meta Llama 3.3 70B Instruct',
+    enabled: true,
+    id: 'Meta-Llama-3.3-70B-Instruct',
+    pricing: {
+      input: 0.6,
+      output: 1.2
+    },
+    type: 'chat'
+  },
+  {
+    contextWindowTokens: 16_000,
+    description: '先进的最尖端小型语言模型，具备语言理解、卓越的推理能力和文本生成能力。',
+    displayName: 'Meta Llama 3.2 1B Instruct',
+    id: 'Meta-Llama-3.2-1B-Instruct',
+    pricing: {
+      input: 0.04,
+      output: 0.08
+    },
+    type: 'chat'
+  },
+  {
+    contextWindowTokens: 8000,
+    description: '先进的最尖端小型语言模型，具备语言理解、卓越的推理能力和文本生成能力。',
+    displayName: 'Meta Llama 3.2 3B Instruct',
+    id: 'Meta-Llama-3.2-3B-Instruct',
+    pricing: {
+      input: 0.08,
+      output: 0.16
+    },
+    type: 'chat'
+  },
+  {
+    abilities: {
+      vision: true,
+    },
+    contextWindowTokens: 4000,
+    description: '在高分辨率图像上表现出色的图像推理能力，适用于视觉理解应用。',
+    displayName: 'Meta Llama 3.2 11B Vision Instruct',
+    enabled: true,
+    id: 'Llama-3.2-11B-Vision-Instruct',
+    pricing: {
+      input: 0.15,
+      output: 0.3
+    },
+    type: 'chat'
+  },
+  {
+    abilities: {
+      vision: true,
+    },
+    contextWindowTokens: 4000,
+    description: '适用于视觉理解代理应用的高级图像推理能力。',
+    displayName: 'Meta Llama 3.2 90B Vision Instruct',
+    enabled: true,
+    id: 'Llama-3.2-90B-Vision-Instruct	',
+    pricing: {
+      input: 0.8,
+      output: 1.6
+    },
+    type: 'chat'
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 16_000,
+    description: 'Llama 3.1指令调优的文本模型，针对多语言对话用例进行了优化，在许多可用的开源和封闭聊天模型中，在常见行业基准上表现优异。',
+    displayName: 'Meta Llama 3.1 8B Instruct',
+    id: 'Meta-Llama-3.1-8B-Instruct',
+    pricing: {
+      input: 0.1,
+      output: 0.2
+    },
+    type: 'chat'
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 128_000,
+    description: 'Llama 3.1指令调优的文本模型，针对多语言对话用例进行了优化，在许多可用的开源和封闭聊天模型中，在常见行业基准上表现优异。',
+    displayName: 'Meta Llama 3.1 70B Instruct',
+    id: 'Meta-Llama-3.1-70B-Instruct',
+    pricing: {
+      input: 0.6,
+      output: 1.2
+    },
+    type: 'chat'
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 16_000,
+    description: 'Llama 3.1指令调优的文本模型，针对多语言对话用例进行了优化，在许多可用的开源和封闭聊天模型中，在常见行业基准上表现优异。',
+    displayName: 'Meta Llama 3.1 405B Instruct',
+    id: 'Meta-Llama-3.1-405B-Instruct',
+    pricing: {
+      input: 5,
+      output: 10
+    },
+    type: 'chat'
+  },
+  {
+    contextWindowTokens: 16_000,
+    displayName: 'Llama 3.1 Tulu 3 405B',
+    id: 'Llama-3.1-Tulu-3-405B',
+    pricing: {
+      input: 0.7,
+      output: 1.4
+    },
+    type: 'chat'
+  },
+  {
+    abilities: {
+      reasoning: true,
+    },
+    contextWindowTokens: 4000,
+    description: '最先进的高效 LLM，擅长推理、数学和编程。',
+    displayName: 'DeepSeek R1',
+    id: 'DeepSeek-R1',
+    pricing: {
+      input: 5,
+      output: 7
+    },
+    type: 'chat'
+  },
+  {
+    abilities: {
+      reasoning: true,
+    },
+    contextWindowTokens: 32_000,
+    description: 'DeepSeek R1——DeepSeek 套件中更大更智能的模型——被蒸馏到 Llama 70B 架构中。基于基准测试和人工评估，该模型比原始 Llama 70B 更智能，尤其在需要数学和事实精确性的任务上表现出色。',
+    displayName: 'DeepSeek R1 Distill Llama 70B',
+    enabled: true,
+    id: 'DeepSeek-R1-Distill-Llama-70B',
+    pricing: {
+      input: 0.7,
+      output: 1.4
+    },
+    type: 'chat'
+  },
+  {
+    abilities: {
+      reasoning: true,
+    },
+    contextWindowTokens: 16_000,
+    description: 'Qwen QwQ 是由 Qwen 团队开发的实验研究模型，专注于提升AI推理能力。',
+    displayName: 'QwQ 32B Preview',
+    enabled: true,
+    id: 'QwQ-32B-Preview',
+    pricing: {
+      input: 1.5,
+      output: 3
+    },
+    type: 'chat'
+  },
+  {
+    contextWindowTokens: 16_000,
+    description: '面向中文和英文的 LLM，针对语言、编程、数学、推理等领域。',
+    displayName: 'Qwen2.5 72B Instruct',
+    enabled: true,
+    id: 'Qwen2.5-72B-Instruct',
+    pricing: {
+      input: 2,
+      output: 4
+    },
+    type: 'chat'
+  },
+  {
+    contextWindowTokens: 16_000,
+    description: '高级 LLM，支持代码生成、推理和修复，涵盖主流编程语言。',
+    displayName: 'Qwen2.5 Coder 32B Instruct',
+    enabled: true,
+    id: 'Qwen2.5-Coder-32B-Instruct',
+    pricing: {
+      input: 1.5,
+      output: 3
+    },
+    type: 'chat'
+  },
+]
+
+export const allModels = [...sambanovaChatModels];
+
+export default allModels;

src/config/llm.ts

@@ -140,6 +140,9 @@ export const getLLMConfig = () => {
 
       ENABLED_JINA: z.boolean(),
       JINA_API_KEY: z.string().optional(),
+
+      ENABLED_SAMBANOVA: z.boolean(),
+      SAMBANOVA_API_KEY: z.string().optional(),
     },
     runtimeEnv: {
       API_KEY_SELECT_MODE: process.env.API_KEY_SELECT_MODE,
@@ -278,6 +281,9 @@ export const getLLMConfig = () => {
 
       ENABLED_JINA: !!process.env.JINA_API_KEY,
       JINA_API_KEY: process.env.JINA_API_KEY,
+
+      ENABLED_SAMBANOVA: !!process.env.SAMBANOVA_API_KEY,
+      SAMBANOVA_API_KEY: process.env.SAMBANOVA_API_KEY,
     },
   });
 };

src/config/modelProviders/index.ts

@@ -31,6 +31,7 @@ import OpenAIProvider from './openai';
 import OpenRouterProvider from './openrouter';
 import PerplexityProvider from './perplexity';
 import QwenProvider from './qwen';
+import SambaNovaProvider from './sambanova';
 import SenseNovaProvider from './sensenova';
 import SiliconCloudProvider from './siliconcloud';
 import SparkProvider from './spark';
@@ -71,6 +72,7 @@ export const LOBE_DEFAULT_MODEL_LIST: ChatModelCard[] = [
   HuggingFaceProvider.chatModels,
   XAIProvider.chatModels,
   JinaProvider.chatModels,
+  SambaNovaProvider.chatModels,
   ZeroOneProvider.chatModels,
   StepfunProvider.chatModels,
   NovitaProvider.chatModels,
@@ -116,6 +118,7 @@ export const DEFAULT_MODEL_PROVIDER_LIST = [
   UpstageProvider,
   XAIProvider,
   JinaProvider,
+  SambaNovaProvider,
   QwenProvider,
   WenxinProvider,
   TencentcloudProvider,
@@ -179,6 +182,7 @@ export { default as OpenAIProviderCard } from './openai';
 export { default as OpenRouterProviderCard } from './openrouter';
 export { default as PerplexityProviderCard } from './perplexity';
 export { default as QwenProviderCard } from './qwen';
+export { default as SambaNovaProviderCard } from './sambanova';
 export { default as SenseNovaProviderCard } from './sensenova';
 export { default as SiliconCloudProviderCard } from './siliconcloud';
 export { default as SparkProviderCard } from './spark';

src/config/modelProviders/sambanova.ts

@@ -0,0 +1,18 @@
+import { ModelProviderCard } from '@/types/llm';
+
+const SambaNova: ModelProviderCard = {
+  chatModels: [],
+  checkModel: 'Meta-Llama-3.2-1B-Instruct',
+  description: 'SambaNova Cloud 可让开发者轻松使用最佳的开源模型，并享受最快的推理速度。',
+  disableBrowserRequest: true,
+  id: 'sambanova',
+  modelsUrl: 'https://cloud.sambanova.ai/plans/pricing',
+  name: 'SambaNova',
+  settings: {
+    disableBrowserRequest: true,
+    sdkType: 'openai',
+  },
+  url: 'https://cloud.sambanova.ai',
+};
+
+export default SambaNova; 

src/libs/agent-runtime/AgentRuntime.ts

@@ -33,6 +33,7 @@ import { LobeOpenAI } from './openai';
 import { LobeOpenRouterAI } from './openrouter';
 import { LobePerplexityAI } from './perplexity';
 import { LobeQwenAI } from './qwen';
+import { LobeSambaNovaAI } from './sambanova';
 import { LobeSenseNovaAI } from './sensenova';
 import { LobeSiliconCloudAI } from './siliconcloud';
 import { LobeSparkAI } from './spark';
@@ -169,6 +170,7 @@ class AgentRuntime {
       openrouter: Partial<ClientOptions>;
       perplexity: Partial<ClientOptions>;
       qwen: Partial<ClientOptions>;
+      sambanova: Partial<ClientOptions>;
       sensenova: Partial<ClientOptions>;
       siliconcloud: Partial<ClientOptions>;
       spark: Partial<ClientOptions>;
@@ -384,6 +386,11 @@ class AgentRuntime {
         break;
       }
 
+      case ModelProvider.SambaNova: {
+        runtimeModel = new LobeSambaNovaAI(params.sambanova);
+        break;
+      }
+
       case ModelProvider.Cloudflare: {
         runtimeModel = new LobeCloudflareAI(params.cloudflare);
         break;

src/libs/agent-runtime/sambanova/index.ts

@@ -0,0 +1,10 @@
+import { ModelProvider } from '../types';
+import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
+
+export const LobeSambaNovaAI = LobeOpenAICompatibleFactory({
+  baseURL: 'https://api.sambanova.ai/v1',
+  debug: {
+    chatCompletion: () => process.env.DEBUG_SAMBANOVA_CHAT_COMPLETION === '1',
+  },
+  provider: ModelProvider.SambaNova,
+});

src/libs/agent-runtime/types/type.ts

@@ -56,6 +56,7 @@ export enum ModelProvider {
   OpenRouter = 'openrouter',
   Perplexity = 'perplexity',
   Qwen = 'qwen',
+  SambaNova = 'sambanova',
   SenseNova = 'sensenova',
   SiliconCloud = 'siliconcloud',
   Spark = 'spark',

src/types/user/settings/keyVaults.ts

@@ -59,6 +59,7 @@ export interface UserKeyVaults {
   password?: string;
   perplexity?: OpenAICompatibleKeyVault;
   qwen?: OpenAICompatibleKeyVault;
+  sambanova?: OpenAICompatibleKeyVault;
   sensenova?: OpenAICompatibleKeyVault;
   siliconcloud?: OpenAICompatibleKeyVault;
   spark?: OpenAICompatibleKeyVault;