@lobehub/chat 1.20.8 → 1.21.1

package/src/libs/agent-runtime/utils/streams/wenxin.test.ts

@@ -0,0 +1,149 @@
+import { describe, expect, it, vi } from 'vitest';
+
+import * as uuidModule from '@/utils/uuid';
+
+import { ChatResp } from '../../wenxin/type';
+import { WenxinResultToStream, WenxinStream } from './wenxin';
+
+const dataStream = [
+  {
+    id: 'as-vb0m37ti8y',
+    object: 'chat.completion',
+    created: 1709089502,
+    sentence_id: 0,
+    is_end: false,
+    is_truncated: false,
+    result: '当然可以，',
+    need_clear_history: false,
+    finish_reason: 'normal',
+    usage: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
+  },
+  {
+    id: 'as-vb0m37ti8y',
+    object: 'chat.completion',
+    created: 1709089504,
+    sentence_id: 1,
+    is_end: false,
+    is_truncated: false,
+    result:
+      '以下是一些建议的自驾游路线，它们涵盖了各种不同的风景和文化体验：\n\n1. **西安-敦煌历史文化之旅**：\n\n\n\t* 路线：西安',
+    need_clear_history: false,
+    finish_reason: 'normal',
+    usage: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
+  },
+  {
+    id: 'as-vb0m37ti8y',
+    object: 'chat.completion',
+    created: 1709089506,
+    sentence_id: 2,
+    is_end: false,
+    is_truncated: false,
+    result: ' - 天水 - 兰州 - 嘉峪关 - 敦煌\n\t* 特点：此路线让您领略到中国西北的丰富历史文化。',
+    need_clear_history: false,
+    finish_reason: 'normal',
+    usage: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
+  },
+  {
+    id: 'as-vb0m37ti8y',
+    object: 'chat.completion',
+    created: 1709089508,
+    sentence_id: 3,
+    is_end: false,
+    is_truncated: false,
+    result: '您可以参观西安的兵马俑、大雁塔，体验兰州的黄河风情，以及在敦煌欣赏壮丽的莫高窟。',
+    need_clear_history: false,
+    finish_reason: 'normal',
+    usage: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
+  },
+  {
+    id: 'as-vb0m37ti8y',
+    object: 'chat.completion',
+    created: 1709089511,
+    sentence_id: 4,
+    is_end: false,
+    is_truncated: false,
+    result: '\n2. **海南环岛热带风情游**：\n\n\n\t* 路线：海口 - 三亚 - 陵水 - 万宁 - 文昌 - 海',
+    need_clear_history: false,
+    finish_reason: 'normal',
+    usage: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
+  },
+  {
+    id: 'as-vb0m37ti8y',
+    object: 'chat.completion',
+    created: 1709089512,
+    sentence_id: 5,
+    is_end: false,
+    is_truncated: false,
+    result:
+      '口\n\t* 特点：海南岛是中国唯一的黎族聚居区，这里有独特的热带风情、美丽的海滩和丰富的水果。',
+    need_clear_history: false,
+    finish_reason: 'normal',
+    usage: { prompt_tokens: 5, completion_tokens: 153, total_tokens: 158 },
+  },
+];
+
+describe('WenxinStream', () => {
+  it('should transform Wenxin stream to protocol stream', async () => {
+    vi.spyOn(uuidModule, 'nanoid').mockReturnValueOnce('1');
+
+    const mockWenxinStream: AsyncIterable<ChatResp> = {
+      // @ts-ignore
+      async *[Symbol.asyncIterator]() {
+        for (const item of dataStream) {
+          yield item;
+        }
+      },
+    };
+
+    const stream = WenxinResultToStream(mockWenxinStream);
+
+    const onStartMock = vi.fn();
+    const onTextMock = vi.fn();
+    const onTokenMock = vi.fn();
+    const onCompletionMock = vi.fn();
+
+    const protocolStream = WenxinStream(stream, {
+      onStart: onStartMock,
+      onText: onTextMock,
+      onToken: onTokenMock,
+      onCompletion: onCompletionMock,
+    });
+
+    const decoder = new TextDecoder();
+    const chunks = [];
+
+    // @ts-ignore
+    for await (const chunk of protocolStream) {
+      chunks.push(decoder.decode(chunk, { stream: true }));
+    }
+
+    expect(chunks).toEqual(
+      [
+        'id: as-vb0m37ti8y',
+        'event: text',
+        `data: "当然可以，"\n`,
+        'id: as-vb0m37ti8y',
+        'event: text',
+        `data: "以下是一些建议的自驾游路线，它们涵盖了各种不同的风景和文化体验：\\n\\n1. **西安-敦煌历史文化之旅**：\\n\\n\\n\\t* 路线：西安"\n`,
+        'id: as-vb0m37ti8y',
+        'event: text',
+        `data: " - 天水 - 兰州 - 嘉峪关 - 敦煌\\n\\t* 特点：此路线让您领略到中国西北的丰富历史文化。"\n`,
+        'id: as-vb0m37ti8y',
+        'event: text',
+        `data: "您可以参观西安的兵马俑、大雁塔，体验兰州的黄河风情，以及在敦煌欣赏壮丽的莫高窟。"\n`,
+        'id: as-vb0m37ti8y',
+        'event: text',
+        `data: "\\n2. **海南环岛热带风情游**：\\n\\n\\n\\t* 路线：海口 - 三亚 - 陵水 - 万宁 - 文昌 - 海"\n`,
+        'id: as-vb0m37ti8y',
+        'event: text',
+        `data: "口\\n\\t* 特点：海南岛是中国唯一的黎族聚居区，这里有独特的热带风情、美丽的海滩和丰富的水果。"\n`,
+      ].map((item) => `${item}\n`),
+    );
+
+    expect(onStartMock).toHaveBeenCalledTimes(1);
+    expect(onTextMock).toHaveBeenNthCalledWith(1, '"当然可以，"');
+    expect(onTextMock).toHaveBeenNthCalledWith(2, '"以下是一些建议的自驾游路线，它们涵盖了各种不同的风景和文化体验：\\n\\n1. **西安-敦煌历史文化之旅**：\\n\\n\\n\\t* 路线：西安"');
+    expect(onTokenMock).toHaveBeenCalledTimes(6);
+    expect(onCompletionMock).toHaveBeenCalledTimes(1);
+  });
+});

package/src/libs/agent-runtime/utils/streams/wenxin.ts

@@ -0,0 +1,46 @@
+import { readableFromAsyncIterable } from 'ai';
+
+import { ChatStreamCallbacks } from '@/libs/agent-runtime';
+import { nanoid } from '@/utils/uuid';
+
+import { ChatResp } from '../../wenxin/type';
+import {
+  StreamProtocolChunk,
+  StreamStack,
+  chatStreamable,
+  createCallbacksTransformer,
+  createSSEProtocolTransformer,
+} from './protocol';
+
+const transformERNIEBotStream = (chunk: ChatResp): StreamProtocolChunk => {
+  const finished = chunk.is_end;
+  if (finished) {
+    return { data: chunk.finish_reason || 'stop', id: chunk.id, type: 'stop' };
+  }
+
+  if (chunk.result) {
+    return { data: chunk.result, id: chunk.id, type: 'text' };
+  }
+
+  return {
+    data: chunk,
+    id: chunk.id,
+    type: 'data',
+  };
+};
+
+export const WenxinResultToStream = (stream: AsyncIterable<ChatResp>) => {
+  // make the response to the streamable format
+  return readableFromAsyncIterable(chatStreamable(stream));
+};
+
+export const WenxinStream = (
+  rawStream: ReadableStream<ChatResp>,
+  callbacks?: ChatStreamCallbacks,
+) => {
+  const streamStack: StreamStack = { id: 'chat_' + nanoid() };
+
+  return rawStream
+    .pipeThrough(createSSEProtocolTransformer(transformERNIEBotStream, streamStack))
+    .pipeThrough(createCallbacksTransformer(callbacks));
+};

package/src/libs/agent-runtime/wenxin/index.ts

@@ -0,0 +1,106 @@
+import { ChatCompletion } from '@baiducloud/qianfan';
+
+// 如果引入了这个类型，那么在跑 type-check 的 tsc 检查中就会抛错，大无语
+// import type QianFanClient from '@baiducloud/qianfan/src/ChatCompletion/index';
+import { safeParseJSON } from '@/utils/safeParseJSON';
+
+import { LobeRuntimeAI } from '../BaseAI';
+import { AgentRuntimeErrorType } from '../error';
+import { ChatCompetitionOptions, ChatStreamPayload } from '../types';
+import { AgentRuntimeError } from '../utils/createError';
+import { debugStream } from '../utils/debugStream';
+import { StreamingResponse } from '../utils/response';
+import { WenxinResultToStream, WenxinStream } from '../utils/streams/wenxin';
+import { ChatResp } from './type';
+
+interface ChatErrorCode {
+  error_code: number;
+  error_msg: string;
+}
+
+export interface LobeWenxinAIParams {
+  accessKey?: string;
+  baseURL?: string;
+  secretKey?: string;
+}
+
+export class LobeWenxinAI implements LobeRuntimeAI {
+  private client: any;
+  baseURL?: string;
+
+  constructor({ accessKey, baseURL, secretKey }: LobeWenxinAIParams = {}) {
+    if (!accessKey || !secretKey)
+      throw AgentRuntimeError.createError(AgentRuntimeErrorType.InvalidProviderAPIKey);
+
+    this.client = new ChatCompletion({
+      QIANFAN_ACCESS_KEY: accessKey,
+      QIANFAN_SECRET_KEY: secretKey,
+    });
+    this.baseURL = baseURL;
+  }
+
+  async chat(payload: ChatStreamPayload, options?: ChatCompetitionOptions) {
+    try {
+      const result = await this.client.chat(
+        { messages: payload.messages as any, stream: true, user_id: options?.user },
+        payload.model,
+      );
+
+      const wenxinStream = WenxinResultToStream(result as AsyncIterable<ChatResp>);
+
+      const [prod, useForDebug] = wenxinStream.tee();
+
+      if (process.env.DEBUG_WENXIN_CHAT_COMPLETION === '1') {
+        debugStream(useForDebug).catch();
+      }
+
+      const stream = WenxinStream(prod, options?.callback);
+
+      // Respond with the stream
+      return StreamingResponse(stream, { headers: options?.headers });
+    } catch (e) {
+      const err = e as Error;
+
+      const error: ChatErrorCode | undefined = safeParseJSON(err.message);
+
+      if (!error) {
+        throw AgentRuntimeError.createError(AgentRuntimeErrorType.AgentRuntimeError, {
+          message: err.message,
+          name: err.name,
+        });
+      }
+
+      // 文心一言错误码
+      // https://cloud.baidu.com/doc/WENXINWORKSHOP/s/tlmyncueh
+      switch (error.error_code) {
+        // Invalid API key or access key
+        case 100:
+        case 13:
+        case 14: {
+          throw AgentRuntimeError.createError(AgentRuntimeErrorType.InvalidProviderAPIKey, error);
+        }
+
+        // quota limit
+        case 4:
+        case 17:
+        case 18:
+        case 19:
+        case 336_501:
+        case 336_502:
+        case 336_503:
+        case 336_504:
+        case 336_505:
+        case 336_507: {
+          throw AgentRuntimeError.createError(AgentRuntimeErrorType.QuotaLimitReached, {
+            errorCode: error.error_code,
+            message: `${error.error_msg} | you can visit https://cloud.baidu.com/doc/WENXINWORKSHOP/s/tlmyncueh for more information about the error code`,
+          });
+        }
+      }
+
+      throw AgentRuntimeError.createError(AgentRuntimeErrorType.ProviderBizError, error);
+    }
+  }
+}
+
+export default LobeWenxinAI;

package/src/libs/agent-runtime/wenxin/type.ts

@@ -0,0 +1,84 @@
+/**
+ * token 用量基类
+ */
+export interface TokenUsage {
+  /**
+   * 回答tokens数
+   */
+  completion_tokens?: number;
+  /**
+   * 问题tokens数
+   */
+  prompt_tokens: number;
+  /**
+   * tokens总数
+   */
+  total_tokens: number;
+}
+
+/**
+ * 响应基类
+ */
+export interface RespBase {
+  /**
+   * 时间戳
+   */
+  created: number;
+  /**
+   * 本轮对话的id
+   */
+  id: string;
+  /**
+   * 表示当前子句是否是最后一句。只有在流式接口模式下会返回该字段
+   */
+  is_end?: boolean;
+  /**
+   * 1：表示输入内容无安全风险
+   * 0：表示输入内容有安全风险
+   */
+  is_safe?: number;
+  /**
+   * 回包类型。
+   *
+   * chat.completion：多轮对话返回
+   */
+  object: string;
+  /**
+   * 对话返回结果
+   */
+  result: string;
+  /**
+   * 表示当前子句的序号。只有在流式接口模式下会返回该字段
+   */
+  sentence_id?: number;
+  /**
+   * token统计信息，token数 = 汉字数+单词数*1.3 （仅为估算逻辑）
+   */
+  usage: TokenUsage;
+}
+
+export interface ChatResp extends RespBase {
+  /**
+   * 当 need_clear_history 为 true 时，此字段会告知第几轮对话有敏感信息，如果是当前问题，ban_round=-1
+   */
+  ban_round: number;
+  /**
+   * 输出内容标识，说明：
+   * · normal：输出内容完全由大模型生成，未触发截断、替换
+   * · stop：输出结果命中入参stop中指定的字段后被截断
+   * · length：达到了最大的token数，根据EB返回结果is_truncated来截断
+   * · content_filter：输出内容被截断、兜底、替换为**等
+   */
+  finish_reason: string;
+  /**
+   * 当前生成的结果是否被截断
+   */
+  is_truncated?: boolean;
+  /**
+   * 表示用户输入是否存在安全，是否关闭当前会话，清理历史会话信息
+   *
+   * true：是，表示用户输入存在安全风险，建议关闭当前会话，清理历史会话信息
+   * false：否，表示用户输入无安全风险
+   */
+  need_clear_history: boolean;
+}

package/src/locales/default/error.ts

@@ -79,6 +79,8 @@ export default {
     InvalidClerkUser: '很抱歉，你当前尚未登录，请先登录或注册账号后继续操作',
     LocationNotSupportError:
       '很抱歉，你的所在地区不支持此模型服务，可能是由于区域限制或服务未开通。请确认当前地区是否支持使用此服务，或尝试使用切换到其他地区后重试。',
+    QuotaLimitReached:
+      '很抱歉，当前 Token 用量或请求次数已达该秘钥的配额(quota)上限，请增加该秘钥的配额或稍后再试',
 
     InvalidProviderAPIKey: '{{provider}} API Key 不正确或为空，请检查 {{provider}} API Key 后重试',
     ProviderBizError: '请求 {{provider}} 服务出错，请根据以下信息排查或重试',

package/src/locales/default/modelProvider.ts

@@ -115,6 +115,26 @@ export default {
       title: '下载指定的 Ollama 模型',
     },
   },
+  wenxin: {
+    accessKey: {
+      desc: '填入百度千帆平台的 Access Key',
+      placeholder: 'Qianfan Access Key',
+      title: 'Access Key',
+    },
+    checker: {
+      desc: '测试 AccessKey / SecretAccess 是否填写正确',
+    },
+    secretKey: {
+      desc: '填入百度千帆平台 Secret Key',
+      placeholder: 'Qianfan Secret Key',
+      title: 'Secret Key',
+    },
+    unlock: {
+      customRegion: '自定义服务区域',
+      description: '输入你的 AccessKey / SecretKey 即可开始会话。应用不会记录你的鉴权配置',
+      title: '使用自定义文心一言鉴权信息',
+    },
+  },
   zeroone: {
     title: '01.AI 零一万物',
   },

package/src/server/globalConfig/index.ts

@@ -52,7 +52,7 @@ export const getServerGlobalConfig = () => {
 
     ENABLED_HUNYUAN,
     HUNYUAN_MODEL_LIST,
-    
+
     ENABLED_DEEPSEEK,
     ENABLED_PERPLEXITY,
     ENABLED_ANTHROPIC,
@@ -96,6 +96,8 @@ export const getServerGlobalConfig = () => {
 
     ENABLED_FIREWORKSAI,
     FIREWORKSAI_MODEL_LIST,
+
+    ENABLED_WENXIN,
   } = getLLMConfig();
 
   const config: GlobalServerConfig = {
@@ -239,6 +241,7 @@ export const getServerGlobalConfig = () => {
         }),
       },
       upstage: { enabled: ENABLED_UPSTAGE },
+      wenxin: { enabled: ENABLED_WENXIN },
       zeroone: {
         enabled: ENABLED_ZEROONE,
         enabledModels: extractEnabledModels(ZEROONE_MODEL_LIST),

package/src/services/_auth.ts

@@ -25,6 +25,20 @@ export const getProviderAuthPayload = (provider: string) => {
       };
     }
 
+    case ModelProvider.Wenxin: {
+      const { secretKey, accessKey } = keyVaultsConfigSelectors.wenxinConfig(
+        useUserStore.getState(),
+      );
+
+      const apiKey = (accessKey || '') + (secretKey || '');
+
+      return {
+        apiKey,
+        wenxinAccessKey: accessKey,
+        wenxinSecretKey: secretKey,
+      };
+    }
+
     case ModelProvider.Azure: {
       const azure = keyVaultsConfigSelectors.azureConfig(useUserStore.getState());
 

package/src/store/user/slices/modelList/selectors/keyVaults.ts

@@ -14,6 +14,7 @@ export const keyVaultsSettings = (s: UserStore): UserKeyVaults =>
 
 const openAIConfig = (s: UserStore) => keyVaultsSettings(s).openai || {};
 const bedrockConfig = (s: UserStore) => keyVaultsSettings(s).bedrock || {};
+const wenxinConfig = (s: UserStore) => keyVaultsSettings(s).wenxin || {};
 const ollamaConfig = (s: UserStore) => keyVaultsSettings(s).ollama || {};
 const azureConfig = (s: UserStore) => keyVaultsSettings(s).azure || {};
 const getVaultByProvider = (provider: GlobalLLMProviderKey) => (s: UserStore) =>
@@ -42,4 +43,5 @@ export const keyVaultsConfigSelectors = {
   ollamaConfig,
   openAIConfig,
   password,
+  wenxinConfig,
 };

package/src/types/user/settings/keyVaults.ts

@@ -16,6 +16,11 @@ export interface AWSBedrockKeyVault {
   sessionToken?: string;
 }
 
+export interface WenxinKeyVault {
+  accessKey?: string;
+  secretKey?: string;
+}
+
 export interface UserKeyVaults {
   ai21?: OpenAICompatibleKeyVault;
   ai360?: OpenAICompatibleKeyVault;
@@ -46,6 +51,7 @@ export interface UserKeyVaults {
   taichu?: OpenAICompatibleKeyVault;
   togetherai?: OpenAICompatibleKeyVault;
   upstage?: OpenAICompatibleKeyVault;
+  wenxin?: WenxinKeyVault;
   zeroone?: OpenAICompatibleKeyVault;
   zhipu?: OpenAICompatibleKeyVault;
 }