Merge pull request #19 from lingmengcan/dev

support ollama locally llm

doc/lingmengcan-ai-202409241125.sql

@@ -220,7 +220,10 @@ INSERT INTO `dict` VALUES
 (102,'CONTROL_NET_TYPE','SoftEdge','SoftEdge (软边缘)',17,0,'SoftEdge (软边缘)','admin','admin','2024-08-12 17:25:47','2024-08-12 17:25:47'),
 (103,'CONTROL_NET_TYPE','SparseCtrl','SparseCtrl (稀疏控制)',18,0,'SparseCtrl (稀疏控制)','admin','admin','2024-08-12 17:26:30','2024-08-12 17:26:30'),
 (104,'CONTROL_NET_TYPE','T2I-Adapter','T2I-Adapter',19,0,'T2I-Adapter','admin','admin','2024-08-12 17:27:08','2024-08-12 17:27:23'),
-(105,'CONTROL_NET_TYPE','Tile','Tile (分块)',20,0,'Tile (分块)','admin','admin','2024-08-12 17:27:09','2024-08-12 17:27:09');
+(105,'CONTROL_NET_TYPE','Tile','Tile (分块)',20,0,'Tile (分块)','admin','admin','2024-08-12 17:27:09','2024-08-12 17:27:09'),
+(106,'0_ROOT_TYPE','LLM_API_TYPE','大语言模型访问类型',9,0,'大语言模型访问类型','admin','admin','2024-09-24 16:53:33','2024-09-24 16:53:40'),
+(107,'LLM_API_TYPE','LLM_API_OPENAI','OpenAI',1,0,'openai','admin','admin','2024-09-24 16:54:46','2024-09-24 16:55:53'),
+(108,'LLM_API_TYPE','LLM_API_OLLAMA','Ollama',2,0,'Ollama','admin','admin','2024-09-24 16:55:13','2024-09-24 16:55:13');
 /*!40000 ALTER TABLE `dict` ENABLE KEYS */;
 UNLOCK TABLES;
 
@@ -305,6 +308,7 @@ CREATE TABLE `llm` (
   `model_type_name` varchar(32) COLLATE utf8mb4_bin NOT NULL,
   `base_url` varchar(512) COLLATE utf8mb4_bin NOT NULL,
   `api_key` varchar(512) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin NOT NULL,
+  `api_type` varchar(32) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin NOT NULL,
   `default_embedding_model` varchar(128) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin NOT NULL,
   `status` tinyint(1) NOT NULL,
   `description` varchar(512) CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci NOT NULL,
@@ -324,6 +328,7 @@ LOCK TABLES `llm` WRITE;
 /*!40000 ALTER TABLE `llm` DISABLE KEYS */;
 INSERT INTO `llm` VALUES
 ('3ada5390-cd0a-4dc5-89f4-a25aba1355a9','hunyuan-13B','GENERAL_LLM','通用模型','http://hunyuanapi.tencent.com/openapi/v1','','',0,'腾讯公司开发的大型语言模型混元大模型（HunYuan）。主要功能是通过丰富的语义理解和计算能力，为用户提供问答式的服务。','admin','admin','2024-06-25 10:49:53','2024-06-25 14:36:06'),
+('51770c03-75dc-4f72-84f8-73e7a1a96739','qwen2.5','GENERAL_LLM','通用模型','','','LLM_API_OLLAMA','bge-large-zh-v1.5',0,'qwen2.5','admin','admin','2024-09-24 16:33:29','2024-09-25 15:46:53'),
 ('601278f2-8bf1-4eff-a684-a7eff61d1274','text-embedding-3-small','EMBEDDING_LLM','Embedding','','','',0,'openai 嵌入式模型','admin','admin','2024-06-11 19:34:12','2024-06-24 20:16:39'),
 ('78e33981-4e01-445d-89e1-bfacc0affb0d','ChatGLM3-6B','GENERAL_LLM','通用模型','http://127.0.0.1:8000/v1/','','bge-large-zh-v1.5',0,'ChatGLM3 是智谱AI和清华大学 KEG 实验室联合发布的对话预训练模型','admin','admin','2024-06-11 16:47:27','2024-06-24 20:16:31'),
 ('86409bd7-666d-4ddb-9809-7c52ef3dc56a','gpt-4','GENERAL_LLM','通用模型','https://oai.hconeai.com/v1','','',0,'GPT-4（Generative Pretrained Transformer 4）是一种自然语言处理（NLP）AI模型。','admin','admin','2024-06-07 17:13:21','2024-06-24 20:16:16'),

service/src/entities/llm.entity.ts

@@ -21,6 +21,9 @@ export class Llm {
   @Column('varchar', { name: 'api_key', length: 128 })
   apiKey: string;
 
+  @Column('varchar', { name: 'api_type', length: 32 })
+  apiType: string;
+
   @Column('varchar', { name: 'default_embedding_model', length: 128 })
   defaultEmbeddingModel: string;
 

service/src/services/chat.service.ts

@@ -17,6 +17,10 @@ import { AIMessage, HumanMessage } from '@langchain/core/messages';
 import { VectorStore } from '@langchain/core/vectorstores';
 import { Chroma } from '@langchain/community/vectorstores/chroma';
 import { LlmService } from './llm.service';
+import { Ollama } from '@langchain/ollama';
+import { Llm } from '@/entities/llm.entity';
+import { ChatPromptValueInterface } from '@langchain/core/dist/prompt_values';
+import { RunnableLike } from '@langchain/core/runnables';
 
 @Injectable()
 export class ChatService {
@@ -82,30 +86,32 @@ export class ChatService {
         },
       );
 
-      return this.chatfileOpenAi(
-        message.messageText,
-        temperature,
-        messageHistory,
-        model.baseUrl,
-        model.apiKey,
-        vectorStore,
-      );
+      return this.chatfileOpenAi(message.messageText, temperature, messageHistory, model, vectorStore);
     } else {
-      return this.chatOpenAi(message.messageText, temperature, messageHistory, model.baseUrl, model.apiKey);
+      return this.chatOpenAi(message.messageText, temperature, messageHistory, model);
     }
   }
 
   //自由对话
-  async chatOpenAi(
-    message: string,
-    temperature: number,
-    messageHistory: ChatMessageHistory,
-    basePath: string,
-    openAIApiKey: string,
-  ) {
+  async chatOpenAi(message: string, temperature: number, messageHistory: ChatMessageHistory, model: Llm) {
     //根据内容回答问题
-    // Instantiate your model and prompt.
-    const llm = new ChatOpenAI({ openAIApiKey, temperature, streaming: true }, { basePath });
+
+    // 工厂函数，创建模型实例
+    function createModelInstance(model: Llm, temperature: number): RunnableLike<ChatPromptValueInterface, unknown> {
+      if (model.apiType === 'LLM_API_OLLAMA') {
+        return new Ollama({
+          model: model.modelName,
+          temperature,
+        });
+      } else {
+        return new ChatOpenAI(
+          { openAIApiKey: model.apiKey, temperature, streaming: true },
+          { basePath: model.baseUrl },
+        );
+      }
+    }
+
+    const llm = createModelInstance(model, temperature);
 
     const prompt = ChatPromptTemplate.fromMessages([
       new MessagesPlaceholder('history'),
@@ -129,15 +135,17 @@ export class ChatService {
     message: string,
     temperature: number,
     messageHistory: ChatMessageHistory,
-    basePath: string,
-    openAIApiKey: string,
+    model: Llm,
     vectorStore: VectorStore,
   ) {
     const result = await vectorStore.similaritySearch(message, 1);
 
     //根据内容回答问题
     // Instantiate your model and prompt.
-    const llm = new ChatOpenAI({ openAIApiKey, temperature, streaming: true }, { basePath });
+    const llm = new ChatOpenAI(
+      { openAIApiKey: model.apiKey, temperature, streaming: true },
+      { basePath: model.baseUrl },
+    );
     const prompt = ChatPromptTemplate.fromMessages([
       SystemMessagePromptTemplate.fromTemplate(
         `基于已知内容, 回答用户问题。如果无法从中得到答案，请说'没有足够的相关信息'已知内容:${result[0].pageContent}`,

service/src/services/llm.service.ts

@@ -137,6 +137,7 @@ export class LlmService {
     entity.modelType = model.modelType;
     entity.baseUrl = model.baseUrl;
     entity.apiKey = model.apiKey;
+    entity.apiType = model.apiType;
     entity.defaultEmbeddingModel = model.defaultEmbeddingModel;
     entity.status = model.status;
     entity.description = model.description;
@@ -161,6 +162,7 @@ export class LlmService {
     entity.modelType = model.modelType;
     entity.baseUrl = model.baseUrl;
     entity.apiKey = model.apiKey;
+    entity.apiType = model.apiType;
     entity.defaultEmbeddingModel = model.defaultEmbeddingModel;
     entity.status = model.status;
     entity.description = model.description ?? '';

web/src/models/llm.ts

@@ -6,6 +6,7 @@ export interface Llm {
   modelTypeName?: string;
   baseUrl?: string;
   apiKey?: string;
+  apiType?: string;
   defaultEmbeddingModel?: string;
   status: string;
   description?: string;

web/src/views/llm/model/index.vue

@@ -83,6 +83,9 @@
             dict-type="LLM_TYPE"
           />
         </n-form-item>
+        <n-form-item label="访问类型OpenAI/Ollama" name="apiType">
+          <selectDict v-model:dict-code="drawerFormData.apiType" dict-type="LLM_API_TYPE" />
+        </n-form-item>
         <n-form-item label="base url" name="baseUrl">
           <n-input v-model:value="drawerFormData.baseUrl" placeholder="请输入模型api url" />
         </n-form-item>
@@ -142,6 +145,7 @@
     modelTypeName: '',
     baseUrl: '',
     apiKey: '',
+    apiType: undefined,
     defaultEmbeddingModel: undefined,
     status: '0',
     description: '',