diff --git a/Dockerfile b/Dockerfile index ae9603a571b2..5b102c2ad16f 100644 --- a/Dockerfile +++ b/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 diff --git a/Dockerfile.database b/Dockerfile.database index 3f1138140b03..c8e033c12664 100644 --- a/Dockerfile.database +++ b/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 diff --git a/src/app/(main)/settings/llm/ProviderList/providers.tsx b/src/app/(main)/settings/llm/ProviderList/providers.tsx index 932992954f7a..8fb6fc359b64 100644 --- a/src/app/(main)/settings/llm/ProviderList/providers.tsx +++ b/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, diff --git a/src/app/api/chat/agentRuntime.ts b/src/app/api/chat/agentRuntime.ts index 5b3f0e98e12c..f215a6ea0004 100644 --- a/src/app/api/chat/agentRuntime.ts +++ b/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 }; } } diff --git a/src/config/llm.ts b/src/config/llm.ts index 1c2219ac2666..f967c108c156 100644 --- a/src/config/llm.ts +++ b/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, }, }); }; diff --git a/src/config/modelProviders/hunyuan.ts b/src/config/modelProviders/hunyuan.ts new file mode 100644 index 000000000000..5499da109cc8 --- /dev/null +++ b/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; diff --git a/src/config/modelProviders/index.ts b/src/config/modelProviders/index.ts index 27094172b378..2dbd8c842b75 100644 --- a/src/config/modelProviders/index.ts +++ b/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'; diff --git a/src/const/settings/llm.ts b/src/const/settings/llm.ts index df2c9cadcf88..77d4eb61441d 100644 --- a/src/const/settings/llm.ts +++ b/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), diff --git a/src/libs/agent-runtime/AgentRuntime.ts b/src/libs/agent-runtime/AgentRuntime.ts index 6a7fedba0329..68ea3ade7cd7 100644 --- a/src/libs/agent-runtime/AgentRuntime.ts +++ b/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; google: { apiKey?: string; baseURL?: string }; groq: Partial; + hunyuan: Partial; minimax: Partial; mistral: Partial; moonshot: Partial; @@ -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); diff --git a/src/libs/agent-runtime/hunyuan/index.test.ts b/src/libs/agent-runtime/hunyuan/index.test.ts new file mode 100644 index 000000000000..8582071760cc --- /dev/null +++ b/src/libs/agent-runtime/hunyuan/index.test.ts @@ -0,0 +1,255 @@ +// @vitest-environment node +import OpenAI from 'openai'; +import { Mock, afterEach, beforeEach, describe, expect, it, vi } from 'vitest'; + +import { + ChatStreamCallbacks, + LobeOpenAICompatibleRuntime, + ModelProvider, +} from '@/libs/agent-runtime'; + +import * as debugStreamModule from '../utils/debugStream'; +import { LobeHunyuanAI } from './index'; + +const provider = ModelProvider.Hunyuan; +const defaultBaseURL = 'https://api.hunyuan.cloud.tencent.com/v1'; + +const bizErrorType = 'ProviderBizError'; +const invalidErrorType = 'InvalidProviderAPIKey'; + +// Mock the console.error to avoid polluting test output +vi.spyOn(console, 'error').mockImplementation(() => {}); + +let instance: LobeOpenAICompatibleRuntime; + +beforeEach(() => { + instance = new LobeHunyuanAI({ apiKey: 'test' }); + + // 使用 vi.spyOn 来模拟 chat.completions.create 方法 + vi.spyOn(instance['client'].chat.completions, 'create').mockResolvedValue( + new ReadableStream() as any, + ); +}); + +afterEach(() => { + vi.clearAllMocks(); +}); + +describe('LobeHunyuanAI', () => { + describe('init', () => { + it('should correctly initialize with an API key', async () => { + const instance = new LobeHunyuanAI({ apiKey: 'test_api_key' }); + expect(instance).toBeInstanceOf(LobeHunyuanAI); + expect(instance.baseURL).toEqual(defaultBaseURL); + }); + }); + + describe('chat', () => { + describe('Error', () => { + it('should return OpenAIBizError with an openai error response when OpenAI.APIError is thrown', async () => { + // Arrange + const apiError = new OpenAI.APIError( + 400, + { + status: 400, + error: { + message: 'Bad Request', + }, + }, + 'Error message', + {}, + ); + + vi.spyOn(instance['client'].chat.completions, 'create').mockRejectedValue(apiError); + + // Act + try { + await instance.chat({ + messages: [{ content: 'Hello', role: 'user' }], + model: 'hunyuan-lite', + temperature: 0, + }); + } catch (e) { + expect(e).toEqual({ + endpoint: defaultBaseURL, + error: { + error: { message: 'Bad Request' }, + status: 400, + }, + errorType: bizErrorType, + provider, + }); + } + }); + + it('should throw AgentRuntimeError with NoOpenAIAPIKey if no apiKey is provided', async () => { + try { + new LobeHunyuanAI({}); + } catch (e) { + expect(e).toEqual({ errorType: invalidErrorType }); + } + }); + + it('should return OpenAIBizError with the cause when OpenAI.APIError is thrown with cause', async () => { + // Arrange + const errorInfo = { + stack: 'abc', + cause: { + message: 'api is undefined', + }, + }; + const apiError = new OpenAI.APIError(400, errorInfo, 'module error', {}); + + vi.spyOn(instance['client'].chat.completions, 'create').mockRejectedValue(apiError); + + // Act + try { + await instance.chat({ + messages: [{ content: 'Hello', role: 'user' }], + model: 'hunyuan-lite', + temperature: 0, + }); + } catch (e) { + expect(e).toEqual({ + endpoint: defaultBaseURL, + error: { + cause: { message: 'api is undefined' }, + stack: 'abc', + }, + errorType: bizErrorType, + provider, + }); + } + }); + + it('should return OpenAIBizError with an cause response with desensitize Url', async () => { + // Arrange + const errorInfo = { + stack: 'abc', + cause: { message: 'api is undefined' }, + }; + const apiError = new OpenAI.APIError(400, errorInfo, 'module error', {}); + + instance = new LobeHunyuanAI({ + apiKey: 'test', + + baseURL: 'https://api.abc.com/v1', + }); + + vi.spyOn(instance['client'].chat.completions, 'create').mockRejectedValue(apiError); + + // Act + try { + await instance.chat({ + messages: [{ content: 'Hello', role: 'user' }], + model: 'hunyuan-lite', + temperature: 0, + }); + } catch (e) { + expect(e).toEqual({ + endpoint: 'https://api.***.com/v1', + error: { + cause: { message: 'api is undefined' }, + stack: 'abc', + }, + errorType: bizErrorType, + provider, + }); + } + }); + + it('should throw an InvalidHunyuanAPIKey error type on 401 status code', async () => { + // Mock the API call to simulate a 401 error + const error = new Error('Unauthorized') as any; + error.status = 401; + vi.mocked(instance['client'].chat.completions.create).mockRejectedValue(error); + + try { + await instance.chat({ + messages: [{ content: 'Hello', role: 'user' }], + model: 'hunyuan-lite', + temperature: 0, + }); + } catch (e) { + // Expect the chat method to throw an error with InvalidHunyuanAPIKey + expect(e).toEqual({ + endpoint: defaultBaseURL, + error: new Error('Unauthorized'), + errorType: invalidErrorType, + provider, + }); + } + }); + + it('should return AgentRuntimeError for non-OpenAI errors', async () => { + // Arrange + const genericError = new Error('Generic Error'); + + vi.spyOn(instance['client'].chat.completions, 'create').mockRejectedValue(genericError); + + // Act + try { + await instance.chat({ + messages: [{ content: 'Hello', role: 'user' }], + model: 'hunyuan-lite', + temperature: 0, + }); + } catch (e) { + expect(e).toEqual({ + endpoint: defaultBaseURL, + errorType: 'AgentRuntimeError', + provider, + error: { + name: genericError.name, + cause: genericError.cause, + message: genericError.message, + stack: genericError.stack, + }, + }); + } + }); + }); + + describe('DEBUG', () => { + it('should call debugStream and return StreamingTextResponse when DEBUG_HUNYUAN_CHAT_COMPLETION is 1', async () => { + // Arrange + const mockProdStream = new ReadableStream() as any; // 模拟的 prod 流 + const mockDebugStream = new ReadableStream({ + start(controller) { + controller.enqueue('Debug stream content'); + controller.close(); + }, + }) as any; + mockDebugStream.toReadableStream = () => mockDebugStream; // 添加 toReadableStream 方法 + + // 模拟 chat.completions.create 返回值，包括模拟的 tee 方法 + (instance['client'].chat.completions.create as Mock).mockResolvedValue({ + tee: () => [mockProdStream, { toReadableStream: () => mockDebugStream }], + }); + + // 保存原始环境变量值 + const originalDebugValue = process.env.DEBUG_HUNYUAN_CHAT_COMPLETION; + + // 模拟环境变量 + process.env.DEBUG_HUNYUAN_CHAT_COMPLETION = '1'; + vi.spyOn(debugStreamModule, 'debugStream').mockImplementation(() => Promise.resolve()); + + // 执行测试 + // 运行你的测试函数，确保它会在条件满足时调用 debugStream + // 假设的测试函数调用，你可能需要根据实际情况调整 + await instance.chat({ + messages: [{ content: 'Hello', role: 'user' }], + model: 'hunyuan-lite', + stream: true, + temperature: 0, + }); + + // 验证 debugStream 被调用 + expect(debugStreamModule.debugStream).toHaveBeenCalled(); + + // 恢复原始环境变量值 + process.env.DEBUG_HUNYUAN_CHAT_COMPLETION = originalDebugValue; + }); + }); + }); +}); diff --git a/src/libs/agent-runtime/hunyuan/index.ts b/src/libs/agent-runtime/hunyuan/index.ts new file mode 100644 index 000000000000..1744ecef4734 --- /dev/null +++ b/src/libs/agent-runtime/hunyuan/index.ts @@ -0,0 +1,10 @@ +import { ModelProvider } from '../types'; +import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory'; + +export const LobeHunyuanAI = LobeOpenAICompatibleFactory({ + baseURL: 'https://api.hunyuan.cloud.tencent.com/v1', + debug: { + chatCompletion: () => process.env.DEBUG_HUNYUAN_CHAT_COMPLETION === '1', + }, + provider: ModelProvider.Hunyuan, +}); diff --git a/src/libs/agent-runtime/types/type.ts b/src/libs/agent-runtime/types/type.ts index ac8e9b2e46bf..ae94893af796 100644 --- a/src/libs/agent-runtime/types/type.ts +++ b/src/libs/agent-runtime/types/type.ts @@ -33,6 +33,7 @@ export enum ModelProvider { Github = 'github', Google = 'google', Groq = 'groq', + Hunyuan = 'hunyuan', Minimax = 'minimax', Mistral = 'mistral', Moonshot = 'moonshot', diff --git a/src/server/globalConfig/index.ts b/src/server/globalConfig/index.ts index 6f03a44fa9d2..c53f7be3ce1b 100644 --- a/src/server/globalConfig/index.ts +++ b/src/server/globalConfig/index.ts @@ -9,6 +9,7 @@ import { GithubProviderCard, GoogleProviderCard, GroqProviderCard, + HunyuanProviderCard, NovitaProviderCard, OllamaProviderCard, OpenAIProviderCard, @@ -49,6 +50,9 @@ export const getServerGlobalConfig = () => { ENABLED_GITHUB, GITHUB_MODEL_LIST, + ENABLED_HUNYUAN, + HUNYUAN_MODEL_LIST, + ENABLED_DEEPSEEK, ENABLED_PERPLEXITY, ENABLED_ANTHROPIC, @@ -160,6 +164,14 @@ export const getServerGlobalConfig = () => { modelString: GROQ_MODEL_LIST, }), }, + hunyuan: { + enabled: ENABLED_HUNYUAN, + enabledModels: extractEnabledModels(HUNYUAN_MODEL_LIST), + serverModelCards: transformToChatModelCards({ + defaultChatModels: HunyuanProviderCard.chatModels, + modelString: HUNYUAN_MODEL_LIST, + }), + }, minimax: { enabled: ENABLED_MINIMAX }, mistral: { enabled: ENABLED_MISTRAL }, moonshot: { enabled: ENABLED_MOONSHOT }, diff --git a/src/types/user/settings/keyVaults.ts b/src/types/user/settings/keyVaults.ts index 8fe21885f606..3a0c6c4f0e4b 100644 --- a/src/types/user/settings/keyVaults.ts +++ b/src/types/user/settings/keyVaults.ts @@ -28,6 +28,7 @@ export interface UserKeyVaults { github?: OpenAICompatibleKeyVault; google?: OpenAICompatibleKeyVault; groq?: OpenAICompatibleKeyVault; + hunyuan?: OpenAICompatibleKeyVault; lobehub?: any; minimax?: OpenAICompatibleKeyVault; mistral?: OpenAICompatibleKeyVault;