memory support oceanbase

docker-setup.sh

@@ -74,6 +74,17 @@ else
     $DOWNLOAD_CMD "data/.config.yaml" "https://raw.githubusercontent.com/xinnan-tech/xiaozhi-esp32-server/main/main/xiaozhi-server/config.yaml"
 fi
 
+# 下载量化模型
+echo "下载量化模型..."
+mkdir -p models/all-MiniLM-L6-v2
+if [ "$DOWNLOAD_CMD" = "powershell -Command Invoke-WebRequest -Uri" ]; then
+    $DOWNLOAD_CMD "https://public.ukp.informatik.tu-darmstadt.de/reimers/sentence-transformers/v0.2/all-MiniLM-L6-v2.zip" $DOWNLOAD_CMD_SUFFIX "models/all-MiniLM-L6-v2/all-MiniLM-L6-v2.zip"
+
+else
+    $DOWNLOAD_CMD "models/all-MiniLM-L6-v2.zip" "https://public.ukp.informatik.tu-darmstadt.de/reimers/sentence-transformers/v0.2/all-MiniLM-L6-v2.zip"
+fi
+unzip models/all-MiniLM-L6-v2.zip -d models/all-MiniLM-L6-v2
+
 # 检查文件是否存在
 echo "检查文件完整性..."
 FILES_TO_CHECK="docker-compose.yml data/.config.yaml models/SenseVoiceSmall/model.pt"

docs/Deployment.md

@@ -306,6 +306,17 @@ LLM:
 - 线路二：百度网盘下载[SenseVoiceSmall](https://pan.baidu.com/share/init?surl=QlgM58FHhYv1tFnUT_A8Sg&pwd=qvna) 提取码:
   `qvna`
 
+如果需要使用 oceanbase 的长期记忆向量化能力需要额外下载量化模型，用于把历史对话向量化到成向量化数据。目前支持 `all-MiniLM-L6-v2` 模型。因为模型较大，需要独立下载，
+
+```
+wget https://public.ukp.informatik.tu-darmstadt.de/reimers/sentence-transformers/v0.2/all-MiniLM-L6-v2.zip
+mkdir models/all-MiniLM-L6-v2
+```
+解压后放在`models/all-MiniLM-L6-v2`目录下。
+
+
+```
+
 ## 运行状态确认
 
 如果你能看到，类似以下日志,则是本项目服务启动成功的标志。

main/xiaozhi-server/config.yaml

@@ -194,6 +194,13 @@ Memory:
   mem_local_short:
     # 本地记忆功能，通过selected_module的llm总结，数据保存在本地，不会上传到服务器
     type: mem_local_short
+  oceanbase:
+    uri: "*****:**"
+    user: "***"
+    password: "***"
+    database: xiaozhi
+    table: xiaozhi
+    model_path: "models/all-MiniLM-L6-v2"
 
 ASR:
   FunASR:

main/xiaozhi-server/core/providers/memory/oceanbase/oceanbase.py

@@ -0,0 +1,92 @@
+import os.path
+import traceback
+from pyobvector import MilvusLikeClient
+from sentence_transformers import SentenceTransformer
+from core.providers.memory.base import MemoryProviderBase ,logger
+
+TAG = __name__
+
+create_table_sql = """
+CREATE TABLE xiaozhi(
+    id INT AUTO_INCREMENT PRIMARY KEY, 
+    role VARCHAR(200),
+    content text,
+    embedding VECTOR(384),
+    VECTOR INDEX idx1(embedding) WITH (distance=L2, type=hnsw)
+    );
+"""
+
+
+class MemoryProvider(MemoryProviderBase):
+    def __init__(self, config):
+        super().__init__(config)
+        self.uri = config.get("uri", "localhost")
+        self.user = config.get("user", "root@test")
+        self.password = config.get("password", "")
+        self.db_name = config.get("database", "xiaozhi")
+        self.table_name = config.get("table_name", "xiaozhi")
+        self.model_path = config.get("model_path", "models/all-MiniLM-L6-v2")
+        self.model_path = os.path.abspath(self.model_path)
+        if not os.path.exists(self.model_path):
+            raise Exception(f"模型路径不存在,请下载量化模型 https://public.ukp.informatik.tu-darmstadt.de/reimers/sentence-transformers/v0.2/all-MiniLM-L6-v2.zip 并解压到: {self.model_path}")
+        logger.bind(tag=TAG).info(f"连接到 oceanbase 服务: {self.uri}")
+        self.client = self.connect_to_client()
+
+    def connect_to_client(self):
+        try:
+            return MilvusLikeClient(uri=self.uri, user=self.user,password=self.password, db_name=self.db_name)
+        except Exception as e:
+            logger.bind(tag=TAG).error(f"连接到 oceanbase 服务时发生错误: {str(e)}")
+            logger.bind(tag=TAG).error(f"请检查配置并确认表是否存在: 初始化sql: {create_table_sql}")
+            logger.bind(tag=TAG).error(f"详细错误: {traceback.format_exc()}")
+            return None
+
+    def _string_to_embeddings(self, sentences):
+        # 加载预训练的 'all-MiniLM-L6-v2' 模型
+        model = SentenceTransformer(self.model_path )
+        embeddings = model.encode(sentences)
+        return embeddings
+
+    def init_memory(self, role_id, llm):
+        super().init_memory(role_id, llm)
+        pass
+
+    async def save_memory(self, msgs):
+        if not self.client or len(msgs) < 2:
+            return None
+
+        try:
+
+            messages =[]
+            for message in msgs:
+                if message.role != "system":
+                    if message.content:
+                        messages.append({"role": message.role, "content": message.content,
+                                        "embedding": self._string_to_embeddings(message.content)})
+
+            for i in range(0, len(messages), 1):
+                self.client.insert(collection_name=self.table_name, data=messages[i:i+1])
+            logger.bind(tag=TAG).info(f"Save memory")
+        except Exception as e:
+            logger.bind(tag=TAG).error(f"保存记忆失败: {traceback.format_exc()}")
+            return None
+
+    async def query_memory(self, query: str) -> str:
+        if not self.client:
+            return ""
+
+        # 把 query 向量化
+        query = self._string_to_embeddings(query)
+
+        try:
+            results = self.client.search(
+                collection_name=self.table_name,
+                data=query,
+                anns_field="embedding",
+                limit=5,
+                output_fields=["role", "content"]
+            )
+            return results
+        except Exception as e:
+            logger.bind(tag=TAG).error(f"查询记忆失败: {traceback.format_exc()}")
+            return ""

main/xiaozhi-server/requirements.txt

@@ -23,5 +23,8 @@ bs4==0.0.2
 modelscope==1.23.2
 sherpa_onnx==1.11.0
 mcp==1.4.1
+pyobvector==0.2.4
 cnlunar==0.2.0
 PySocks==1.7.1
+sentence-transformers==4.0.1
+transformers==4.50.3