@lobehub/chat 1.45.16 → 1.46.0

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

@@ -1,184 +1,40 @@
-import { isEmpty } from 'lodash-es';
-import OpenAI from 'openai';
-
-import { LobeRuntimeAI } from '../BaseAI';
-import { AgentRuntimeErrorType } from '../error';
-import {
-  ChatCompetitionOptions,
-  ChatCompletionErrorPayload,
-  ChatStreamPayload,
-  ModelProvider,
-} from '../types';
-import { AgentRuntimeError } from '../utils/createError';
-import { debugStream } from '../utils/debugStream';
-import { StreamingResponse } from '../utils/response';
-import { MinimaxStream } from '../utils/streams';
-
-interface MinimaxBaseResponse {
-  base_resp?: {
-    status_code?: number;
-    status_msg?: string;
-  };
-}
-
-type MinimaxResponse = Partial<OpenAI.ChatCompletionChunk> & MinimaxBaseResponse;
-
-function throwIfErrorResponse(data: MinimaxResponse) {
-  // error status code
-  // https://www.minimaxi.com/document/guides/chat-model/pro/api?id=6569c85948bc7b684b30377e#3.1.3%20%E8%BF%94%E5%9B%9E(response)%E5%8F%82%E6%95%B0
-  if (!data.base_resp?.status_code || data.base_resp?.status_code < 1000) {
-    return;
-  }
-  if (data.base_resp?.status_code === 1004) {
-    throw AgentRuntimeError.chat({
-      error: {
-        code: data.base_resp.status_code,
-        message: data.base_resp.status_msg,
-      },
-      errorType: AgentRuntimeErrorType.InvalidProviderAPIKey,
-      provider: ModelProvider.Minimax,
-    });
-  }
-  throw AgentRuntimeError.chat({
-    error: {
-      code: data.base_resp.status_code,
-      message: data.base_resp.status_msg,
-    },
-    errorType: AgentRuntimeErrorType.ProviderBizError,
-    provider: ModelProvider.Minimax,
-  });
-}
-
-function parseMinimaxResponse(chunk: string): MinimaxResponse | undefined {
-  let body = chunk;
-  if (body.startsWith('data:')) {
-    body = body.slice(5).trim();
-  }
-  if (isEmpty(body)) {
-    return;
-  }
-  return JSON.parse(body) as MinimaxResponse;
-}
-
-export class LobeMinimaxAI implements LobeRuntimeAI {
-  apiKey: string;
-
-  constructor({ apiKey }: { apiKey?: string } = {}) {
-    if (!apiKey) throw AgentRuntimeError.createError(AgentRuntimeErrorType.InvalidProviderAPIKey);
-
-    this.apiKey = apiKey;
-  }
-
-  async chat(payload: ChatStreamPayload, options?: ChatCompetitionOptions): Promise<Response> {
-    try {
-      const response = await fetch('https://api.minimax.chat/v1/text/chatcompletion_v2', {
-        body: JSON.stringify(this.buildCompletionsParams(payload)),
-        headers: {
-          'Authorization': `Bearer ${this.apiKey}`,
-          'Content-Type': 'application/json',
-        },
-        method: 'POST',
-      });
-      if (!response.body || !response.ok) {
-        throw AgentRuntimeError.chat({
-          error: {
-            status: response.status,
-            statusText: response.statusText,
+import { ModelProvider } from '../types';
+import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
+
+import Minimax from '@/config/modelProviders/minimax';
+
+export const getMinimaxMaxOutputs = (modelId: string): number | undefined => {
+  const model = Minimax.chatModels.find(model => model.id === modelId);
+  return model ? model.maxOutput : undefined;
+};
+
+export const LobeMinimaxAI = LobeOpenAICompatibleFactory({
+  baseURL: 'https://api.minimax.chat/v1',
+  chatCompletion: {
+    handlePayload: (payload) => {
+      const { temperature, top_p, ...params } = payload;
+
+      return {
+        ...params,
+        frequency_penalty: undefined,
+        max_tokens: payload.max_tokens !== undefined ? payload.max_tokens : getMinimaxMaxOutputs(payload.model),
+        presence_penalty: undefined,
+        stream: true,
+        temperature: temperature === undefined || temperature <= 0 ? undefined : temperature / 2,
+        tools: params.tools?.map((tool) => ({
+          function: {
+            description: tool.function.description,
+            name: tool.function.name,
+            parameters: JSON.stringify(tool.function.parameters),
           },
-          errorType: AgentRuntimeErrorType.ProviderBizError,
-          provider: ModelProvider.Minimax,
-        });
-      }
-
-      const [prod, body2] = response.body.tee();
-      const [prod2, debug] = body2.tee();
-
-      if (process.env.DEBUG_MINIMAX_CHAT_COMPLETION === '1') {
-        debugStream(debug).catch(console.error);
-      }
-
-      // wait for the first response, and throw error if minix returns an error
-      await this.parseFirstResponse(prod2.getReader());
-
-      return StreamingResponse(MinimaxStream(prod), { headers: options?.headers });
-    } catch (error) {
-      console.log('error', error);
-      const err = error as Error | ChatCompletionErrorPayload;
-      if ('provider' in err) {
-        throw error;
-      }
-      const errorResult = {
-        cause: err.cause,
-        message: err.message,
-        name: err.name,
-        stack: err.stack,
-      };
-      throw AgentRuntimeError.chat({
-        error: errorResult,
-        errorType: AgentRuntimeErrorType.ProviderBizError,
-        provider: ModelProvider.Minimax,
-      });
-    }
-  }
-
-  // the document gives the default value of max tokens, but abab6.5 and abab6.5s
-  // will meet length finished error, and output is truncationed
-  // so here fill the max tokens number to fix it
-  // https://www.minimaxi.com/document/guides/chat-model/V2
-  private getMaxTokens(model: string): number | undefined {
-    switch (model) {
-      case 'abab6.5t-chat':
-      case 'abab6.5g-chat':
-      case 'abab5.5s-chat':
-      case 'abab5.5-chat': {
-        return 4096;
-      }
-      case 'abab6.5s-chat': {
-        return 8192;
-      }
-    }
-  }
-
-  private buildCompletionsParams(payload: ChatStreamPayload) {
-    const { temperature, top_p, ...params } = payload;
-
-    return {
-      ...params,
-      frequency_penalty: undefined,
-      max_tokens:
-        payload.max_tokens !== undefined ? payload.max_tokens : this.getMaxTokens(payload.model),
-      presence_penalty: undefined,
-      stream: true,
-      temperature: temperature === undefined || temperature <= 0 ? undefined : temperature / 2,
-
-      tools: params.tools?.map((tool) => ({
-        function: {
-          description: tool.function.description,
-          name: tool.function.name,
-          parameters: JSON.stringify(tool.function.parameters),
-        },
-        type: 'function',
-      })),
-      top_p: top_p === 0 ? undefined : top_p,
-    };
-  }
-
-  private async parseFirstResponse(reader: ReadableStreamDefaultReader<Uint8Array>) {
-    const decoder = new TextDecoder();
-
-    const { value } = await reader.read();
-    const chunkValue = decoder.decode(value, { stream: true });
-    let data;
-    try {
-      data = parseMinimaxResponse(chunkValue);
-    } catch {
-      // parse error, skip it
-      return;
-    }
-    if (data) {
-      throwIfErrorResponse(data);
-    }
-  }
-}
-
-export default LobeMinimaxAI;
+          type: 'function',
+        })),
+        top_p: top_p !== undefined && top_p > 0 && top_p <= 1 ? top_p : undefined,
+      } as any;
+    },
+  },
+  debug: {
+    chatCompletion: () => process.env.DEBUG_MINIMAX_CHAT_COMPLETION === '1',
+  },
+  provider: ModelProvider.Minimax,
+});

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

@@ -6,7 +6,13 @@ import { ChatModelCard } from '@/types/llm';
 
 import { LobeRuntimeAI } from '../BaseAI';
 import { AgentRuntimeErrorType } from '../error';
-import { ChatCompetitionOptions, ChatStreamPayload, ModelProvider } from '../types';
+import {
+  ChatCompetitionOptions,
+  ChatStreamPayload,
+  Embeddings,
+  EmbeddingsPayload,
+  ModelProvider,
+} from '../types';
 import { AgentRuntimeError } from '../utils/createError';
 import { debugStream } from '../utils/debugStream';
 import { StreamingResponse } from '../utils/response';
@@ -84,6 +90,18 @@ export class LobeOllamaAI implements LobeRuntimeAI {
     }
   }
 
+  async embeddings(payload: EmbeddingsPayload): Promise<Embeddings[]> {
+    const input = Array.isArray(payload.input) ? payload.input : [payload.input];
+    const promises = input.map((inputText: string) =>
+      this.invokeEmbeddingModel({
+        dimensions: payload.dimensions,
+        input: inputText,
+        model: payload.model,
+      }),
+    );
+    return await Promise.all(promises);
+  }
+
   async models(): Promise<ChatModelCard[]> {
     const list = await this.client.list();
     return list.models.map((model) => ({
@@ -91,6 +109,24 @@ export class LobeOllamaAI implements LobeRuntimeAI {
     }));
   }
 
+  private invokeEmbeddingModel = async (payload: EmbeddingsPayload): Promise<Embeddings> => {
+    try {
+      const responseBody = await this.client.embeddings({
+        model: payload.model,
+        prompt: payload.input as string,
+      });
+      return responseBody.embedding;
+    } catch (error) {
+      const e = error as { message: string; name: string; status_code: number };
+
+      throw AgentRuntimeError.chat({
+        error: { message: e.message, name: e.name, status_code: e.status_code },
+        errorType: AgentRuntimeErrorType.OllamaBizError,
+        provider: ModelProvider.Ollama,
+      });
+    }
+  };
+
   private buildOllamaMessages(messages: OpenAIChatMessage[]) {
     return messages.map((message) => this.convertContentToOllamaMessage(message));
   }

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

@@ -39,6 +39,7 @@ export enum ModelProvider {
   HuggingFace = 'huggingface',
   Hunyuan = 'hunyuan',
   InternLM = 'internlm',
+  LMStudio = 'lmstudio',
   Minimax = 'minimax',
   Mistral = 'mistral',
   Moonshot = 'moonshot',

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

@@ -2,7 +2,6 @@ export * from './anthropic';
 export * from './azureOpenai';
 export * from './bedrock';
 export * from './google-ai';
-export * from './minimax';
 export * from './ollama';
 export * from './openai';
 export * from './protocol';

package/src/server/globalConfig/index.ts

@@ -1,6 +1,7 @@
 import { appEnv, getAppConfig } from '@/config/app';
 import { authEnv } from '@/config/auth';
 import { fileEnv } from '@/config/file';
+import { knowledgeEnv } from '@/config/knowledge';
 import { langfuseEnv } from '@/config/langfuse';
 import { enableNextAuth } from '@/const/auth';
 import { parseSystemAgent } from '@/server/globalConfig/parseSystemAgent';
@@ -9,6 +10,7 @@ import { GlobalServerConfig } from '@/types/serverConfig';
 import { genServerLLMConfig } from './_deprecated';
 import { genServerAiProvidersConfig } from './genServerAiProviderConfig';
 import { parseAgentConfig } from './parseDefaultAgent';
+import { parseFilesConfig } from './parseFilesConfig';
 
 export const getServerGlobalConfig = () => {
   const { ACCESS_CODES, DEFAULT_AGENT_CONFIG } = getAppConfig();
@@ -73,3 +75,7 @@ export const getServerDefaultAgentConfig = () => {
 
   return parseAgentConfig(DEFAULT_AGENT_CONFIG) || {};
 };
+
+export const getServerDefaultFilesConfig = () => {
+  return parseFilesConfig(knowledgeEnv.DEFAULT_FILES_CONFIG);
+};

package/src/server/globalConfig/parseFilesConfig.test.ts

@@ -0,0 +1,17 @@
+import { describe, expect, it } from 'vitest';
+
+import { parseFilesConfig } from './parseFilesConfig';
+
+describe('parseFilesConfig', () => {
+  // 测试embeddings配置是否被正确解析
+  it('parses embeddings configuration correctly', () => {
+    const envStr =
+      'embedding_model=openai/embedding-text-3-large,reranker_model=cohere/rerank-english-v3.0,query_model=full_text';
+    const expected = {
+      embeddingModel: { provider: 'openai', model: 'embedding-text-3-large' },
+      rerankerModel: { provider: 'cohere', model: 'rerank-english-v3.0' },
+      queryModel: 'full_text',
+    };
+    expect(parseFilesConfig(envStr)).toEqual(expected);
+  });
+});

package/src/server/globalConfig/parseFilesConfig.ts

@@ -0,0 +1,57 @@
+import { DEFAULT_FILES_CONFIG } from '@/const/settings/knowledge';
+import { SystemEmbeddingConfig } from '@/types/knowledgeBase';
+import { FilesConfig } from '@/types/user/settings/filesConfig';
+
+const protectedKeys = Object.keys({
+  embedding_model: null,
+  query_model: null,
+  reranker_model: null,
+});
+
+export const parseFilesConfig = (envString: string = ''): SystemEmbeddingConfig => {
+  if (!envString) return DEFAULT_FILES_CONFIG;
+  const config: FilesConfig = {} as any;
+
+  // 处理全角逗号和多余空格
+  let envValue = envString.replaceAll('，', ',').trim();
+
+  const pairs = envValue.split(',');
+
+  for (const pair of pairs) {
+    const [key, value] = pair.split('=').map((s) => s.trim());
+
+    if (key && value) {
+      const [provider, ...modelParts] = value.split('/');
+      const model = modelParts.join('/');
+
+      if ((!provider || !model) && key !== 'query_model') {
+        throw new Error('Missing model or provider value');
+      }
+
+      if (key === 'query_model' && value === '') {
+        throw new Error('Missing query mode value');
+      }
+
+      if (protectedKeys.includes(key)) {
+        switch (key) {
+          case 'embedding_model': {
+            config.embeddingModel = { model: model.trim(), provider: provider.trim() };
+            break;
+          }
+          case 'reranker_model': {
+            config.rerankerModel = { model: model.trim(), provider: provider.trim() };
+            break;
+          }
+          case 'query_model': {
+            config.queryModel = value;
+            break;
+          }
+        }
+      }
+    } else {
+      throw new Error('Invalid environment variable format');
+    }
+  }
+
+  return config;
+};

package/src/server/routers/async/file.ts

@@ -5,15 +5,15 @@ import { z } from 'zod';
 
 import { serverDBEnv } from '@/config/db';
 import { fileEnv } from '@/config/file';
-import { DEFAULT_EMBEDDING_MODEL } from '@/const/settings';
+import { DEFAULT_FILE_EMBEDDING_MODEL_ITEM } from '@/const/settings/knowledge';
 import { NewChunkItem, NewEmbeddingsItem } from '@/database/schemas';
 import { serverDB } from '@/database/server';
 import { ASYNC_TASK_TIMEOUT, AsyncTaskModel } from '@/database/server/models/asyncTask';
 import { ChunkModel } from '@/database/server/models/chunk';
 import { EmbeddingModel } from '@/database/server/models/embedding';
 import { FileModel } from '@/database/server/models/file';
-import { ModelProvider } from '@/libs/agent-runtime';
 import { asyncAuthedProcedure, asyncRouter as router } from '@/libs/trpc/async';
+import { getServerDefaultFilesConfig } from '@/server/globalConfig';
 import { initAgentRuntimeWithUserPayload } from '@/server/modules/AgentRuntime';
 import { S3 } from '@/server/modules/S3';
 import { ChunkService } from '@/server/services/chunk';
@@ -44,7 +44,6 @@ export const fileRouter = router({
     .input(
       z.object({
         fileId: z.string(),
-        model: z.string().default(DEFAULT_EMBEDDING_MODEL),
         taskId: z.string(),
       }),
     )
@@ -57,6 +56,9 @@ export const fileRouter = router({
 
       const asyncTask = await ctx.asyncTaskModel.findById(input.taskId);
 
+      const { model, provider } =
+        getServerDefaultFilesConfig().embeddingModel || DEFAULT_FILE_EMBEDDING_MODEL_ITEM;
+
       if (!asyncTask) throw new TRPCError({ code: 'BAD_REQUEST', message: 'Async Task not found' });
 
       try {
@@ -84,13 +86,12 @@ export const fileRouter = router({
 
           const chunks = await ctx.chunkModel.getChunksTextByFileId(input.fileId);
           const requestArray = chunk(chunks, CHUNK_SIZE);
-
           try {
             await pMap(
               requestArray,
               async (chunks, index) => {
                 const agentRuntime = await initAgentRuntimeWithUserPayload(
-                  ModelProvider.OpenAI,
+                  provider,
                   ctx.jwtPayload,
                 );
 
@@ -98,11 +99,10 @@ export const fileRouter = router({
                 console.log(`执行第 ${number} 个任务`);
 
                 console.time(`任务[${number}]: embeddings`);
-
                 const embeddings = await agentRuntime.embeddings({
                   dimensions: 1024,
                   input: chunks.map((c) => c.text),
-                  model: input.model,
+                  model,
                 });
                 console.timeEnd(`任务[${number}]: embeddings`);
 
@@ -111,7 +111,7 @@ export const fileRouter = router({
                     chunkId: chunks[idx].id,
                     embeddings: e,
                     fileId: input.fileId,
-                    model: input.model,
+                    model,
                   })) || [];
 
                 console.time(`任务[${number}]: insert db`);

package/src/server/routers/lambda/chunk.ts

@@ -1,7 +1,7 @@
 import { inArray } from 'drizzle-orm/expressions';
 import { z } from 'zod';
 
-import { DEFAULT_EMBEDDING_MODEL } from '@/const/settings';
+import { DEFAULT_FILE_EMBEDDING_MODEL_ITEM } from '@/const/settings/knowledge';
 import { knowledgeBaseFiles } from '@/database/schemas';
 import { serverDB } from '@/database/server';
 import { AsyncTaskModel } from '@/database/server/models/asyncTask';
@@ -9,9 +9,9 @@ import { ChunkModel } from '@/database/server/models/chunk';
 import { EmbeddingModel } from '@/database/server/models/embedding';
 import { FileModel } from '@/database/server/models/file';
 import { MessageModel } from '@/database/server/models/message';
-import { ModelProvider } from '@/libs/agent-runtime';
 import { authedProcedure, router } from '@/libs/trpc';
 import { keyVaults } from '@/libs/trpc/middleware/keyVaults';
+import { getServerDefaultFilesConfig } from '@/server/globalConfig';
 import { initAgentRuntimeWithUserPayload } from '@/server/modules/AgentRuntime';
 import { ChunkService } from '@/server/services/chunk';
 import { SemanticSearchSchema } from '@/types/rag';
@@ -101,21 +101,18 @@ export const chunkRouter = router({
     .input(
       z.object({
         fileIds: z.array(z.string()).optional(),
-        model: z.string().default(DEFAULT_EMBEDDING_MODEL),
         query: z.string(),
       }),
     )
     .mutation(async ({ ctx, input }) => {
-      console.time('embedding');
-      const agentRuntime = await initAgentRuntimeWithUserPayload(
-        ModelProvider.OpenAI,
-        ctx.jwtPayload,
-      );
+      const { model, provider } =
+        getServerDefaultFilesConfig().embeddingModel || DEFAULT_FILE_EMBEDDING_MODEL_ITEM;
+      const agentRuntime = await initAgentRuntimeWithUserPayload(provider, ctx.jwtPayload);
 
       const embeddings = await agentRuntime.embeddings({
         dimensions: 1024,
         input: input.query,
-        model: input.model,
+        model,
       });
       console.timeEnd('embedding');
 
@@ -130,27 +127,25 @@ export const chunkRouter = router({
     .input(SemanticSearchSchema)
     .mutation(async ({ ctx, input }) => {
       const item = await ctx.messageModel.findMessageQueriesById(input.messageId);
+      const { model, provider } =
+        getServerDefaultFilesConfig().embeddingModel || DEFAULT_FILE_EMBEDDING_MODEL_ITEM;
       let embedding: number[];
       let ragQueryId: string;
-
       // if there is no message rag or it's embeddings, then we need to create one
       if (!item || !item.embeddings) {
         // TODO: need to support customize
-        const agentRuntime = await initAgentRuntimeWithUserPayload(
-          ModelProvider.OpenAI,
-          ctx.jwtPayload,
-        );
+        const agentRuntime = await initAgentRuntimeWithUserPayload(provider, ctx.jwtPayload);
 
         const embeddings = await agentRuntime.embeddings({
           dimensions: 1024,
           input: input.rewriteQuery,
-          model: input.model || DEFAULT_EMBEDDING_MODEL,
+          model,
         });
 
         embedding = embeddings![0];
         const embeddingsId = await ctx.embeddingModel.create({
           embeddings: embedding,
-          model: input.model,
+          model,
         });
 
         const result = await ctx.messageModel.createMessageQuery({
@@ -182,6 +177,7 @@ export const chunkRouter = router({
         fileIds: finalFileIds,
         query: input.rewriteQuery,
       });
+      // TODO: need to rerank the chunks
       console.timeEnd('semanticSearch');
 
       return { chunks, queryId: ragQueryId };

package/src/types/knowledgeBase/index.ts

@@ -1,3 +1,5 @@
+import { FilesConfigItem } from '../user/settings/filesConfig';
+
 export enum KnowledgeBaseTabs {
   Files = 'files',
   Settings = 'Settings',
@@ -43,3 +45,9 @@ export interface KnowledgeItem {
   name: string;
   type: KnowledgeType;
 }
+
+export interface SystemEmbeddingConfig {
+  embeddingModel: FilesConfigItem;
+  queryModel: string;
+  rerankerModel: FilesConfigItem;
+}

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

@@ -0,0 +1,9 @@
+export interface FilesConfigItem {
+  model: string;
+  provider: string;
+}
+export interface FilesConfig {
+  embeddingModel: FilesConfigItem;
+  queryModel: string;
+  rerankerModel: FilesConfigItem;
+}

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

@@ -44,6 +44,7 @@ export interface UserKeyVaults {
   huggingface?: OpenAICompatibleKeyVault;
   hunyuan?: OpenAICompatibleKeyVault;
   internlm?: OpenAICompatibleKeyVault;
+  lmstudio?: OpenAICompatibleKeyVault;
   lobehub?: any;
   minimax?: OpenAICompatibleKeyVault;
   mistral?: OpenAICompatibleKeyVault;

package/src/app/(backend)/webapi/chat/minimax/route.test.ts

@@ -1,26 +0,0 @@
-// @vitest-environment edge-runtime
-import { describe, expect, it, vi } from 'vitest';
-
-import { POST as UniverseRoute } from '../[provider]/route';
-import { POST, runtime } from './route';
-
-// 模拟 '../[provider]/route'
-vi.mock('../[provider]/route', () => ({
-  POST: vi.fn().mockResolvedValue('mocked response'),
-}));
-
-describe('Configuration tests', () => {
-  it('should have runtime set to "edge"', () => {
-    expect(runtime).toBe('nodejs');
-  });
-});
-
-describe('Minimax POST function tests', () => {
-  it('should call UniverseRoute with correct parameters', async () => {
-    const mockRequest = new Request('https://example.com', { method: 'POST' });
-    await POST(mockRequest);
-    expect(UniverseRoute).toHaveBeenCalledWith(mockRequest, {
-      params: Promise.resolve({ provider: 'minimax' }),
-    });
-  });
-});

package/src/app/(backend)/webapi/chat/minimax/route.ts

@@ -1,6 +0,0 @@
-import { POST as UniverseRoute } from '../[provider]/route';
-
-export const runtime = 'nodejs';
-
-export const POST = async (req: Request) =>
-  UniverseRoute(req, { params: Promise.resolve({ provider: 'minimax' }) });