feat(lightrag): 添加LightRAG知识库的语言选择和LLM模型选择功能

- 支持在创建LightRAG知识库时选择语言和LLM模型 - 修复知识图谱加载按钮在没有文件时仍可点击的问题 - 更新README中的分支说明 - 简化chroma和milvus的集合命名方式 - 修复知识库创建失败的情况
2025-08-16 20:07:12 +08:00 · 2025-08-16 20:07:12 +08:00 · e6145657c5
commit e6145657c5
parent 9e4402aa60
12 changed files with 110 additions and 36 deletions
--- a/README.md
+++ b/README.md
@ -37,10 +37,10 @@ https://github.com/user-attachments/assets/15f7f315-003d-4e41-a260-739c2529f824

 1. **克隆项目**
   ```bash
-   git clone -b stable https://github.com/xerrors/Yuxi-Know.git
+   git clone -b 0.2.0.preview https://github.com/xerrors/Yuxi-Know.git
   cd Yuxi-Know
   ```
-   如果想要使用 v0.2 预览版，可以使用分支：`0.2.0.preview`
+   如果想要使用之前的稳定版，可以使用分支：`stable` 分支，`main` 分支是最新的开发版本。

 2. **配置 API 密钥**

--- a/docs/changelog/update.md
+++ b/docs/changelog/update.md
@ -19,7 +19,8 @@
 - [x] 切换知识库之后，检索结果没有刷新
 - [x] 文件上传模块 UI的边距、配色有问题
 - [x] 知识图谱页面的节点数量统计方法 #236
- [ ] 当没有手动添加节点的时候，尝试检索，会出现为创建索引的情况 #236
+- [x] 当没有手动添加节点的时候，尝试检索，会出现为创建索引的情况 #236
+- [ ] 删除知识库的时候，没有正常将数据从 Neo4j / Milvus 数据库中删除（需要添加检测脚本）

 # 💯 More:

--- a/server/routers/knowledge_router.py
+++ b/server/routers/knowledge_router.py
@ -33,10 +33,11 @@ async def create_database(
    embed_model_name: str = Body(...),
    kb_type: str = Body("lightrag"),
    additional_params: dict = Body({}),
+    llm_info: dict = Body(None),
    current_user: User = Depends(get_admin_user)
 ):
    """创建知识库"""
-    logger.debug(f"Create database {database_name} with kb_type {kb_type}, additional_params {additional_params}")
+    logger.debug(f"Create database {database_name} with kb_type {kb_type}, additional_params {additional_params}, llm_info {llm_info}")
    try:
        embed_info = config.embed_model_names[embed_model_name]
        database_info = await knowledge_base.create_database(
@ -44,6 +45,7 @@ async def create_database(
            description,
            kb_type=kb_type,
            embed_info=embed_info,
+            llm_info=llm_info,
            **additional_params
        )

--- a/src/knowledge/chroma_kb.py
+++ b/src/knowledge/chroma_kb.py
@ -64,7 +64,7 @@ class ChromaKB(KnowledgeBase):
        embedding_function = self._get_embedding_function(embed_info)

        # 创建或获取集合
-        collection_name = f"kb_{db_id}"
+        collection_name = db_id

        try:
            # 尝试获取现有集合
--- a/src/knowledge/knowledge_base.py
+++ b/src/knowledge/knowledge_base.py
@ -85,7 +85,7 @@ class KnowledgeBase(ABC):
        pass

    def create_database(self, database_name: str, description: str,
-                       embed_info: dict | None = None, **kwargs) -> dict:
+                       embed_info: dict | None = None, llm_info: dict | None = None, **kwargs) -> dict:
        """
        创建数据库

@ -108,6 +108,7 @@ class KnowledgeBase(ABC):
            "description": description,
            "kb_type": self.kb_type,
            "embed_info": embed_info,
+            "llm_info": llm_info,
            "metadata": kwargs,
            "created_at": datetime.now().isoformat()
        }
@ -187,7 +188,6 @@ class KnowledgeBase(ABC):
        """
        pass

-    @abstractmethod
    async def export_data(self, db_id: str, format: str = 'zip', **kwargs) -> str:
        pass

--- a/src/knowledge/lightrag_kb.py
+++ b/src/knowledge/lightrag_kb.py
@ -55,6 +55,16 @@ class LightRagKB(KnowledgeBase):

        llm_info = self.databases_meta[db_id].get("llm_info", {})
        embed_info = self.databases_meta[db_id].get("embed_info", {})
+        # 读取在创建数据库时透传的附加参数（包括语言）
+        metadata = self.databases_meta[db_id].get("metadata", {}) or {}
+        addon_params = {}
+        if isinstance(metadata.get("addon_params"), dict):
+            addon_params.update(metadata.get("addon_params", {}))
+        # 兼容直接放在 metadata 下的 language
+        if isinstance(metadata.get("language"), str) and metadata.get("language"):
+            addon_params.setdefault("language", metadata.get("language"))
+        # 默认语言从环境变量读取，默认 English
+        addon_params.setdefault("language", os.getenv("SUMMARY_LANGUAGE", "English"))

        # 创建工作目录
        working_dir = os.path.join(self.work_dir, db_id)
@ -71,6 +81,7 @@ class LightRagKB(KnowledgeBase):
            graph_storage="Neo4JStorage",
            doc_status_storage="JsonDocStatusStorage",
            log_file_path=os.path.join(working_dir, "lightrag.log"),
+            addon_params=addon_params,
        )

        return rag
@ -107,7 +118,18 @@ class LightRagKB(KnowledgeBase):
    def _get_llm_func(self, llm_info: dict):
        """获取 LLM 函数"""
        from src.models import select_model
-        model = select_model(LIGHTRAG_LLM_PROVIDER, LIGHTRAG_LLM_NAME)
+
+        # 如果用户选择了LLM，使用用户选择的；否则使用环境变量默认值
+        if llm_info and llm_info.get("provider") and llm_info.get("model_name"):
+            provider = llm_info["provider"]
+            model_name = llm_info["model_name"]
+            logger.info(f"Using user-selected LLM: {provider}/{model_name}")
+        else:
+            provider = LIGHTRAG_LLM_PROVIDER
+            model_name = LIGHTRAG_LLM_NAME
+            logger.info(f"Using default LLM from environment: {provider}/{model_name}")
+
+        model = select_model(provider, model_name)

        async def llm_model_func(prompt, system_prompt=None, history_messages=[], **kwargs):
            return await openai_complete_if_cache(
--- a/src/knowledge/milvus_kb.py
+++ b/src/knowledge/milvus_kb.py
@ -101,7 +101,7 @@ class MilvusKB(KnowledgeBase):
            raise ValueError(f"Database {db_id} not found")

        embed_info = self.databases_meta[db_id].get("embed_info", {})
-        collection_name = f"kb_{db_id}"
+        collection_name = db_id

        try:
            # 检查集合是否存在
@ -256,7 +256,7 @@ class MilvusKB(KnowledgeBase):
                self._save_metadata()

            file_record["file_id"] = file_id
-            
+
            # 添加到处理队列
            self._add_to_processing_queue(file_id)

--- a/test/data/lightrag_kb_test_tiny.txt
+++ b/test/data/lightrag_kb_test_tiny.txt
@ -0,0 +1 @@
+《红楼梦》是中国古典四大名著之一，由曹雪芹创作
--- a/web/src/components/DebugComponent.vue
+++ b/web/src/components/DebugComponent.vue
@ -370,10 +370,7 @@ const printDatabaseInfo = async () => {
  if (!checkAdminPermission()) return;

  try {
-    console.log('=== 知识库信息 ===');
-
-    // 直接调用API获取最新的数据库信息
-    await databaseStore.refreshDatabase();
+    console.log('知识库信息', databaseStore.database);

  } catch (error) {
    console.error('获取知识库信息失败:', error);
--- a/web/src/components/KnowledgeGraphSection.vue
+++ b/web/src/components/KnowledgeGraphSection.vue
@ -15,6 +15,7 @@
          size="small"
          @click="loadGraph"
          :disabled="!isGraphSupported"
+          :icon='h(ReloadOutlined)'
        >
          加载图谱
        </a-button>
@ -153,7 +154,6 @@
 import { ref, computed, watch } from 'vue';
 import { useDatabaseStore } from '@/stores/database';
 import { useUserStore } from '@/stores/user';
-import { getKbTypeLabel } from '@/utils/kb_utils';
 import { ReloadOutlined, DeleteOutlined, ExpandOutlined, UpOutlined, DownOutlined, SettingOutlined } from '@ant-design/icons-vue';
 import { message } from 'ant-design-vue';
 import KnowledgeGraphViewer from '@/components/KnowledgeGraphViewer.vue';
@ -208,6 +208,9 @@ const toggleVisible = () => {
 };

 const loadGraph = () => {
+  if (!(Object.keys(store.database?.files).length > 0)) {
+    return;
+  }
  if (graphViewerRef.value && typeof graphViewerRef.value.loadFullGraph === 'function') {
    graphViewerRef.value.loadFullGraph();
  }
--- a/web/src/stores/database.js
+++ b/web/src/stores/database.js
@ -241,17 +241,17 @@ export const useDatabaseStore = defineStore('database', () => {
    try {
      const response = await queryApi.getKnowledgeBaseQueryParams(db_id);
      queryParams.value = response.params?.options || [];
-      
+
      // Create a set of currently supported parameter keys
      const supportedParamKeys = new Set(queryParams.value.map(param => param.key));
-      
+
      // Remove unsupported parameters from meta
      for (const key in meta) {
        if (key !== 'db_id' && !supportedParamKeys.has(key)) {
          delete meta[key];
        }
      }
-      
+
      // Add default values for supported parameters that are not in meta
      queryParams.value.forEach(param => {
        if (!(param.key in meta)) {
@ -291,16 +291,16 @@ export const useDatabaseStore = defineStore('database', () => {
      stopAutoRefresh();
    }
  }
-  
+
  function selectAllFailedFiles() {
    const files = Object.values(database.value.files || {});
    const failedFiles = files
      .filter(file => file.status === 'failed')
      .map(file => file.file_id);
-  
+
    const newSelectedKeys = [...new Set([...selectedRowKeys.value, ...failedFiles])];
    selectedRowKeys.value = newSelectedKeys;
-  
+
    if (failedFiles.length > 0) {
      message.success(`已选择 ${failedFiles.length} 个失败的文件`);
    } else {
--- a/web/src/views/DataBaseView.vue
+++ b/web/src/views/DataBaseView.vue
@ -47,18 +47,26 @@
      <h3>嵌入模型</h3>
      <a-select v-model:value="newDatabase.embed_model_name" :options="embedModelOptions" style="width: 100%;" size="large" />

-      <!-- 根据类型显示不同配置 -->
-      <!-- <div v-if="newDatabase.kb_type === 'chroma' || newDatabase.kb_type === 'milvus'" class="storage-config">
-        <h3>存储配置</h3>
-        <div class="param-row">
-          <label>存储方式：</label>
-          <a-select v-model:value="newDatabase.storage" style="width: 200px;">
-            <a-select-option value="DemoA">DemoA</a-select-option>
-            <a-select-option value="DemoB">DemoB</a-select-option>
-          </a-select>
-          <span class="param-hint">存储方式配置（功能预留）</span>
-        </div>
-      </div> -->
+      <!-- 仅对 LightRAG 提供语言选择和LLM选择 -->
+      <div v-if="newDatabase.kb_type === 'lightrag'">
+        <h3 style="margin-top: 20px;">语言</h3>
+        <a-select
+          v-model:value="newDatabase.language"
+          :options="languageOptions"
+          style="width: 100%;"
+          size="large"
+          :dropdown-match-select-width="false"
+        />
+
+        <h3 style="margin-top: 20px;">语言模型 (LLM)</h3>
+        <p style="color: var(--gray-700); font-size: 14px;">可以在设置中配置语言模型</p>
+        <ModelSelectorComponent
+          :model_name="newDatabase.llm_info.model_name || '请选择模型'"
+          :model_provider="newDatabase.llm_info.provider || ''"
+          @select-model="handleLLMSelect"
+          style="width: 100%; height: 60px;"
+        />
+      </div>

      <h3 style="margin-top: 20px;">知识库描述</h3>
      <p style="color: var(--gray-700); font-size: 14px;">在智能体流程中，这里的描述会作为工具的描述。智能体会根据知识库的标题和描述来选择合适的工具。所以这里描述的越详细，智能体越容易选择到合适的工具。</p>
@ -72,13 +80,13 @@
        <a-button key="submit" type="primary" :loading="state.creating" @click="createDatabase">创建</a-button>
      </template>
    </a-modal>
-    
+
    <!-- 加载状态 -->
    <div v-if="state.loading" class="loading-container">
      <a-spin size="large" />
      <p>正在加载知识库...</p>
    </div>
-    
+
    <!-- 数据库列表 -->
    <div v-else class="databases">
      <div class="new-database dbcard" @click="state.openNewDatabaseModel=true">
@ -115,7 +123,7 @@
        <p class="description">{{ database.description || '暂无描述' }}</p>
        <div class="tags">
          <a-tag color="blue" v-if="database.embed_info?.name">{{ database.embed_info.name }}</a-tag>
-          <a-tag color="green" v-if="database.embed_info?.dimension">{{ database.embed_info.dimension }}</a-tag>
+          <!-- <a-tag color="green" v-if="database.embed_info?.dimension">{{ database.embed_info.dimension }}</a-tag> -->
          <a-tag
            :color="getKbTypeColor(database.kb_type || 'lightrag')"
            class="kb-type-tag"
@ -139,6 +147,7 @@ import { message } from 'ant-design-vue'
 import { BookPlus, Database, Zap, FileDigit,  Waypoints, Building2 } from 'lucide-vue-next';
 import { databaseApi, typeApi } from '@/apis/knowledge_api';
 import HeaderComponent from '@/components/HeaderComponent.vue';
+import ModelSelectorComponent from '@/components/ModelSelectorComponent.vue';

 const route = useRoute()
 const router = useRouter()
@ -158,6 +167,21 @@ const embedModelOptions = computed(() => {
  }))
 })

+// 语言选项（值使用英文，以保证后端/LightRAG 兼容；标签为中英文方便理解）
+const languageOptions = [
+  { label: '英语 English', value: 'English' },
+  { label: '中文 Chinese', value: 'Chinese' },
+  { label: '日语 Japanese', value: 'Japanese' },
+  { label: '韩语 Korean', value: 'Korean' },
+  { label: '德语 German', value: 'German' },
+  { label: '法语 French', value: 'French' },
+  { label: '西班牙语 Spanish', value: 'Spanish' },
+  { label: '葡萄牙语 Portuguese', value: 'Portuguese' },
+  { label: '俄语 Russian', value: 'Russian' },
+  { label: '阿拉伯语 Arabic', value: 'Arabic' },
+  { label: '印地语 Hindi', value: 'Hindi' },
+]
+
 const emptyEmbedInfo = {
  name: '',
  description: '',
@ -165,6 +189,12 @@ const emptyEmbedInfo = {
  kb_type: 'chroma', // 默认为 Milvus
  // Vector 知识库特有配置
  storage: '', // 存储方式配置
+  // LightRAG 特有配置
+  language: 'English',
+  llm_info: {
+    provider: '',
+    model_name: ''
+  },
 }

 const newDatabase = reactive({
@ -312,6 +342,13 @@ const handleKbTypeChange = (type) => {
  newDatabase.kb_type = type
 }

+// 处理LLM选择
+const handleLLMSelect = (selection) => {
+  console.log('LLM选择:', selection)
+  newDatabase.llm_info.provider = selection.provider
+  newDatabase.llm_info.model_name = selection.name
+}
+
 const createDatabase = () => {
  if (!newDatabase.name?.trim()) {
    message.error('数据库名称不能为空')
@ -338,6 +375,17 @@ const createDatabase = () => {
    requestData.additional_params.storage = newDatabase.storage || 'DemoA'
  }

+  if (newDatabase.kb_type === 'lightrag') {
+    requestData.additional_params.language = newDatabase.language || 'English'
+    // 添加LLM信息到请求数据
+    if (newDatabase.llm_info.provider && newDatabase.llm_info.model_name) {
+      requestData.llm_info = {
+        provider: newDatabase.llm_info.provider,
+        model_name: newDatabase.llm_info.model_name
+      }
+    }
+  }
+
  databaseApi.createDatabase(requestData)
    .then(data => {
      console.log('创建成功:', data)
				`@ -0,0 +1 @@`
				`《红楼梦》是中国古典四大名著之一，由曹雪芹创作`