feat(similarity): 添加异步和批量相似度计算功能，优化性能

feat(graph_store): 增强图存储管理，添加边的注册和注销功能
feat(memory_tools): 支持批量生成嵌入向量
feat(unified_manager): 优化感知记忆和短期记忆的检索逻辑

src/memory_graph/perceptual_manager.py

@@ -21,7 +21,7 @@ import numpy as np
 from src.common.logger import get_logger
 from src.memory_graph.models import MemoryBlock, PerceptualMemory
 from src.memory_graph.utils.embeddings import EmbeddingGenerator
-from src.memory_graph.utils.similarity import cosine_similarity
+from src.memory_graph.utils.similarity import cosine_similarity_async, batch_cosine_similarity_async
 
 logger = get_logger(__name__)
 
@@ -430,14 +430,22 @@ class PerceptualMemoryManager:
                 logger.warning("查询向量生成失败，返回空列表")
                 return []
 
-            # 计算所有块的相似度
+            # 批量计算所有块的相似度（使用异步版本）
+            blocks_with_embeddings = [
+                block for block in self.perceptual_memory.blocks
+                if block.embedding is not None
+            ]
+
+            if not blocks_with_embeddings:
+                return []
+
+            # 批量计算相似度
+            block_embeddings = [block.embedding for block in blocks_with_embeddings]
+            similarities = await batch_cosine_similarity_async(query_embedding, block_embeddings)
+
+            # 过滤和排序
             scored_blocks = []
-            for block in self.perceptual_memory.blocks:
+            for block, similarity in zip(blocks_with_embeddings, similarities):
-                if block.embedding is None:
-                    continue
-
-                similarity = cosine_similarity(query_embedding, block.embedding)
-
                 # 过滤低于阈值的块
                 if similarity >= similarity_threshold:
                     scored_blocks.append((block, similarity))

src/memory_graph/short_term_manager.py

@@ -25,7 +25,7 @@ from src.memory_graph.models import (
     ShortTermOperation,
 )
 from src.memory_graph.utils.embeddings import EmbeddingGenerator
-from src.memory_graph.utils.similarity import cosine_similarity
+from src.memory_graph.utils.similarity import cosine_similarity_async, batch_cosine_similarity_async
 
 logger = get_logger(__name__)
 
@@ -457,7 +457,7 @@ class ShortTermMemoryManager:
                 if existing_mem.embedding is None:
                     continue
 
-                similarity = cosine_similarity(memory.embedding, existing_mem.embedding)
+                similarity = await cosine_similarity_async(memory.embedding, existing_mem.embedding)
                 scored.append((existing_mem, similarity))
 
             # 按相似度降序排序
@@ -567,7 +567,7 @@ class ShortTermMemoryManager:
                 if memory.embedding is None:
                     continue
 
-                similarity = cosine_similarity(query_embedding, memory.embedding)
+                similarity = await cosine_similarity_async(query_embedding, memory.embedding)
                 if similarity >= similarity_threshold:
                     scored.append((memory, similarity))
 

src/memory_graph/storage/graph_store.py

@@ -4,6 +4,8 @@
 
 from __future__ import annotations
 
+from collections.abc import Iterable
+
 import networkx as nx
 
 from src.common.logger import get_logger
@@ -34,8 +36,50 @@ class GraphStore:
         # 索引：节点ID -> 所属记忆ID集合
         self.node_to_memories: dict[str, set[str]] = {}
         
+        # 节点 -> {memory_id: [MemoryEdge]}，用于快速获取邻接边
+        self.node_edge_index: dict[str, dict[str, list[MemoryEdge]]] = {}
+
         logger.info("初始化图存储")
 
+
+    def _register_memory_edges(self, memory: Memory) -> None:
+        """在记忆中的边加入邻接索引"""
+        for edge in memory.edges:
+            self._register_edge_reference(memory.id, edge)
+
+    def _register_edge_reference(self, memory_id: str, edge: MemoryEdge) -> None:
+        """在节点邻接索引中登记一条边"""
+        for node_id in (edge.source_id, edge.target_id):
+            node_edges = self.node_edge_index.setdefault(node_id, {})
+            edge_list = node_edges.setdefault(memory_id, [])
+            if not any(existing.id == edge.id for existing in edge_list):
+                edge_list.append(edge)
+
+    def _unregister_memory_edges(self, memory: Memory) -> None:
+        """从节点邻接索引中移除记忆相关的边"""
+        for edge in memory.edges:
+            self._unregister_edge_reference(memory.id, edge)
+
+    def _unregister_edge_reference(self, memory_id: str, edge: MemoryEdge) -> None:
+        """在节点邻接索引中删除一条边"""
+        for node_id in (edge.source_id, edge.target_id):
+            node_edges = self.node_edge_index.get(node_id)
+            if not node_edges:
+                continue
+            if memory_id not in node_edges:
+                continue
+            node_edges[memory_id] = [e for e in node_edges[memory_id] if e.id != edge.id]
+            if not node_edges[memory_id]:
+                del node_edges[memory_id]
+            if not node_edges:
+                del self.node_edge_index[node_id]
+
+    def _rebuild_node_edge_index(self) -> None:
+        """重建节点邻接索引"""
+        self.node_edge_index.clear()
+        for memory in self.memory_index.values():
+            self._register_memory_edges(memory)
+
     def add_memory(self, memory: Memory) -> None:
         """
         添加记忆到图
@@ -77,6 +121,9 @@ class GraphStore:
             # 3. 保存记忆对象
             self.memory_index[memory.id] = memory
 
+            # 4. 注册记忆中的边到邻接索引
+            self._register_memory_edges(memory)
+
             logger.debug(f"添加记忆到图: {memory}")
 
         except Exception as e:
@@ -112,6 +159,12 @@ class GraphStore:
 
             memory = self.memory_index[memory_id]
 
+            # 1.5. 注销记忆中的边的邻接索引记录
+            self._unregister_memory_edges(memory)
+
+            # 1.5. 注销记忆中的边的邻接索引记录
+            self._unregister_memory_edges(memory)
+
             # 2. 添加节点到图
             if not self.graph.has_node(node_id):
                 from datetime import datetime
@@ -282,6 +335,7 @@ class GraphStore:
                 memory = self.memory_index.get(mem_id)
                 if memory:
                     memory.edges.append(new_edge)
+                    self._register_edge_reference(mem_id, new_edge)
 
             logger.debug(f"添加边成功: {source_id} -> {target_id} ({relation})")
             return edge_id
@@ -393,6 +447,10 @@ class GraphStore:
             for mem_id in related_memory_ids:
                 memory = self.memory_index.get(mem_id)
                 if memory:
+                    removed_edges = [e for e in memory.edges if e.id == edge_id]
+                    if removed_edges:
+                        for edge_obj in removed_edges:
+                            self._unregister_edge_reference(mem_id, edge_obj)
                     memory.edges = [e for e in memory.edges if e.id != edge_id]
 
             return True
@@ -440,8 +498,11 @@ class GraphStore:
                 # 2. 转移边
                 for edge in source_memory.edges:
                     # 添加到目标记忆（如果不存在）
-                    if not any(e.id == edge.id for e in target_memory.edges):
+                    already_exists = any(e.id == edge.id for e in target_memory.edges)
+                    self._unregister_edge_reference(source_id, edge)
+                    if not already_exists:
                         target_memory.edges.append(edge)
+                        self._register_edge_reference(target_memory_id, edge)
 
                 # 3. 删除源记忆（不清理孤立节点，因为节点已转移）
                 del self.memory_index[source_id]
@@ -465,6 +526,32 @@ class GraphStore:
         """
         return self.memory_index.get(memory_id)
 
+    def get_memories_by_ids(self, memory_ids: Iterable[str]) -> dict[str, Memory]:
+        """
+        根据一组ID批量获取记忆
+
+        Args:
+            memory_ids: 记忆ID集合
+
+        Returns:
+            {memory_id: Memory} 映射
+        """
+        result: dict[str, Memory] = {}
+        missing_ids: list[str] = []
+
+        # dict.fromkeys 可以保持参数序列的原始顺序同时帮忙去重
+        for mem_id in dict.fromkeys(memory_ids):
+            memory = self.memory_index.get(mem_id)
+            if memory is not None:
+                result[mem_id] = memory
+            else:
+                missing_ids.append(mem_id)
+
+        if missing_ids:
+            logger.debug(f"批量获取记忆: 未找到 {missing_ids}")
+
+        return result
+
     def get_all_memories(self) -> list[Memory]:
         """
         获取所有记忆
@@ -490,6 +577,32 @@ class GraphStore:
         memory_ids = self.node_to_memories[node_id]
         return [self.memory_index[mid] for mid in memory_ids if mid in self.memory_index]
 
+    def get_edges_for_node(self, node_id: str) -> list[MemoryEdge]:
+        """
+        获取节点相关的全部边（包含入边和出边）
+
+        Args:
+            node_id: 节点ID
+
+        Returns:
+            MemoryEdge 列表
+        """
+        node_edges = self.node_edge_index.get(node_id)
+        if not node_edges:
+            return []
+
+        unique_edges: dict[str | tuple[str, str, str, str], MemoryEdge] = {}
+        for edges in node_edges.values():
+            for edge in edges:
+                key: str | tuple[str, str, str, str]
+                if edge.id:
+                    key = edge.id
+                else:
+                    key = (edge.source_id, edge.target_id, edge.relation, edge.edge_type.value)
+                unique_edges[key] = edge
+
+        return list(unique_edges.values())
+
     def get_edges_from_node(self, node_id: str, relation_types: list[str] | None = None) -> list[dict]:
         """
         获取从指定节点出发的所有边
@@ -762,6 +875,8 @@ class GraphStore:
         except Exception:
             logger.exception("同步图边到记忆.edges 失败")
 
+        store._rebuild_node_edge_index()
+
         logger.info(f"从字典加载图: {store.get_statistics()}")
         return store
 
@@ -829,6 +944,7 @@ class GraphStore:
                 existing_edges.setdefault(mid, set()).add(mem_edge.id)
 
         logger.info("已将图中的边同步到 Memory.edges（保证 graph 与 memory 对象一致）")
+        self._rebuild_node_edge_index()
 
     def remove_memory(self, memory_id: str, cleanup_orphans: bool = True) -> bool:
         """
@@ -877,4 +993,5 @@ class GraphStore:
         self.graph.clear()
         self.memory_index.clear()
         self.node_to_memories.clear()
+        self.node_edge_index.clear()
         logger.warning("图存储已清空")

src/memory_graph/tools/memory_tools.py

@@ -1171,8 +1171,10 @@ class MemoryTools:
             query_embeddings = []
             query_weights = []
 
-            for sub_query, weight in multi_queries:
+            batch_texts = [sub_query for sub_query, _ in multi_queries]
-                embedding = await self.builder.embedding_generator.generate(sub_query)
+            batch_embeddings = await self.builder.embedding_generator.generate_batch(batch_texts)
+
+            for (sub_query, weight), embedding in zip(multi_queries, batch_embeddings):
                 if embedding is not None:
                     query_embeddings.append(embedding)
                     query_weights.append(weight)

src/memory_graph/unified_manager.py

@@ -228,8 +228,13 @@ class UnifiedMemoryManager:
             }
 
             # 步骤1: 检索感知记忆和短期记忆
-            perceptual_blocks = await self.perceptual_manager.recall_blocks(query_text)
+            perceptual_blocks_task = asyncio.create_task(self.perceptual_manager.recall_blocks(query_text))
-            short_term_memories = await self.short_term_manager.search_memories(query_text)
+            short_term_memories_task = asyncio.create_task(self.short_term_manager.search_memories(query_text))
+
+            perceptual_blocks, short_term_memories = await asyncio.gather(
+                perceptual_blocks_task,
+                short_term_memories_task,
+            )
             
             # 步骤1.5: 检查需要转移的感知块，推迟到后台处理
             blocks_to_transfer = [

src/memory_graph/utils/__init__.py

@@ -4,7 +4,12 @@
 
 from src.memory_graph.utils.embeddings import EmbeddingGenerator, get_embedding_generator
 from src.memory_graph.utils.path_expansion import Path, PathExpansionConfig, PathScoreExpansion
-from src.memory_graph.utils.similarity import cosine_similarity
+from src.memory_graph.utils.similarity import (
+    cosine_similarity,
+    cosine_similarity_async,
+    batch_cosine_similarity,
+    batch_cosine_similarity_async
+)
 from src.memory_graph.utils.time_parser import TimeParser
 
 __all__ = [
@@ -14,5 +19,8 @@ __all__ = [
     "PathScoreExpansion",
     "TimeParser",
     "cosine_similarity",
+    "cosine_similarity_async",
+    "batch_cosine_similarity",
+    "batch_cosine_similarity_async",
     "get_embedding_generator",
 ]

src/memory_graph/utils/embeddings.py

@@ -137,56 +137,69 @@ class EmbeddingGenerator:
 
         raise ValueError("无法确定嵌入向量维度，请确保已正确配置 embedding API")
 
+
     async def generate_batch(self, texts: list[str]) -> list[np.ndarray | None]:
-        """
+        """保留输入顺序的批量嵌入生成"""
-        批量生成嵌入向量
-
-        Args:
-            texts: 文本列表
-
-        Returns:
-            嵌入向量列表，失败的项目为 None
-        """
         if not texts:
             return []
 
         try:
-            # 过滤空文本
+            results: list[np.ndarray | None] = [None] * len(texts)
-            valid_texts = [t for t in texts if t and t.strip()]
+            valid_entries = [
-            if not valid_texts:
+                (idx, text) for idx, text in enumerate(texts) if text and text.strip()
-                logger.debug("所有文本为空，返回 None 列表")
+            ]
-                return [None for _ in texts]
+            if not valid_entries:
-
+                logger.debug('批量文本为空，返回空列表')
-            # 使用 API 批量生成（如果可用）
-            if self.use_api:
-                results = await self._generate_batch_with_api(valid_texts)
-                if results:
                 return results
 
-            # 回退到逐个生成
+            batch_texts = [text for _, text in valid_entries]
-            results = []
+            batch_embeddings: list[np.ndarray | None] | None = None
-            for text in valid_texts:
+
-                embedding = await self.generate(text)
+            if self.use_api:
-                results.append(embedding)
+                batch_embeddings = await self._generate_batch_with_api(batch_texts)
+
+            if not batch_embeddings:
+                batch_embeddings = []
+                for _, text in valid_entries:
+                    batch_embeddings.append(await self.generate(text))
+
+            for (idx, _), embedding in zip(valid_entries, batch_embeddings):
+                results[idx] = embedding
 
             success_count = sum(1 for r in results if r is not None)
-            logger.debug(f"✅ 批量生成嵌入: {success_count}/{len(texts)} 个成功")
+            logger.debug(f"批量生成嵌入: {success_count}/{len(texts)}")
             return results
 
         except Exception as e:
-            logger.error(f"❌ 批量嵌入生成失败: {e}", exc_info=True)
+            logger.error(f"批量生成嵌入失败: {e}", exc_info=True)
             return [None for _ in texts]
 
     async def _generate_batch_with_api(self, texts: list[str]) -> list[np.ndarray | None] | None:
-        """使用 API 批量生成"""
+        """使用嵌入 API 在单次请求中生成向量"""
+        if not texts:
+            return []
+
         try:
-            # 对于大多数 API，批量调用就是多次单独调用
+            if not self._api_available:
-            # 这里保持简单，逐个调用
+                await self._initialize_api()
-            results = []
+
-            for text in texts:
+            if not self._api_available or not self._llm_request:
-                embedding = await self._generate_with_api(text)
+                return None
-                results.append(embedding)  # 失败的项目为 None，不中断整个批量处理
+
+            embeddings, model_name = await self._llm_request.get_embedding(texts)
+            if not embeddings:
+                return None
+
+            results: list[np.ndarray | None] = []
+            for emb in embeddings:
+                if emb:
+                    results.append(np.array(emb, dtype=np.float32))
+                else:
+                    results.append(None)
+
+            logger.debug(f"API 批量生成 {len(texts)} 个嵌入向量，使用模型: {model_name}")
             return results
+
         except Exception as e:
             logger.debug(f"API 批量生成失败: {e}")
             return None

src/memory_graph/utils/path_expansion.py

@@ -15,13 +15,14 @@
 """
 
 import asyncio
+import heapq
 import time
 from dataclasses import dataclass, field
 from datetime import datetime, timezone
 from typing import TYPE_CHECKING, Any
 
 from src.common.logger import get_logger
-from src.memory_graph.utils.similarity import cosine_similarity
+from src.memory_graph.utils.similarity import cosine_similarity_async
 
 if TYPE_CHECKING:
     import numpy as np
@@ -273,9 +274,8 @@ class PathScoreExpansion:
                     f"⚠️  路径数量超限 ({len(next_paths)} > {self.config.max_active_paths})，"
                     f"保留 top {self.config.top_paths_retain}"
                 )
-                next_paths = sorted(next_paths, key=lambda p: p.score, reverse=True)[
+                retain = min(self.config.top_paths_retain, len(next_paths))
-                    : self.config.top_paths_retain
+                next_paths = heapq.nlargest(retain, next_paths, key=lambda p: p.score)
-                ]
 
             # 🚀 早停检测：如果路径增长很少，提前终止
             prev_path_count = len(active_paths)
@@ -398,22 +398,14 @@ class PathScoreExpansion:
         if node_id in self._neighbor_cache:
             return self._neighbor_cache[node_id]
 
-        edges = []
+        edges = self.graph_store.get_edges_for_node(node_id)
 
-        # 从图存储中获取与该节点相关的所有边
+        if not edges:
-        # 需要遍历所有记忆找到包含该节点的边
+            self._neighbor_cache[node_id] = []
-        for memory_id in self.graph_store.node_to_memories.get(node_id, []):
+            return []
-            memory = self.graph_store.get_memory_by_id(memory_id)
-            if memory:
-                for edge in memory.edges:
-                    if edge.source_id == node_id or edge.target_id == node_id:
-                        edges.append(edge)
-
-        # 去重（同一条边可能出现多次）
-        unique_edges = list({(e.source_id, e.target_id, e.edge_type): e for e in edges}.values())
 
         # 按边权重排序
-        unique_edges.sort(key=lambda e: self._get_edge_weight(e), reverse=True)
+        unique_edges = sorted(edges, key=lambda e: self._get_edge_weight(e), reverse=True)
 
         # 🚀 存入缓存
         self._neighbor_cache[node_id] = unique_edges
@@ -461,7 +453,7 @@ class PathScoreExpansion:
                 base_score = 0.3  # 无向量的节点给低分
             else:
                 node_embedding = node_data["embedding"]
-                similarity = cosine_similarity(query_embedding, node_embedding)
+                similarity = await cosine_similarity_async(query_embedding, node_embedding)
                 base_score = max(0.0, min(1.0, similarity))  # 限制在[0, 1]
 
         # 🆕 偏好类型加成
@@ -522,14 +514,8 @@ class PathScoreExpansion:
                 node_metadata_map[nid] = node_data.get("metadata", {})
 
         if valid_embeddings:
-            # 批量计算相似度（使用矩阵运算）
+            # 批量计算相似度（使用矩阵运算）- 移至to_thread执行
-            embeddings_matrix = np.array(valid_embeddings)
+            similarities = await asyncio.to_thread(self._batch_compute_similarities, valid_embeddings, query_embedding)
-            query_norm = np.linalg.norm(query_embedding)
-            embeddings_norms = np.linalg.norm(embeddings_matrix, axis=1)
-
-            # 向量化计算余弦相似度
-            similarities = np.dot(embeddings_matrix, query_embedding) / (embeddings_norms * query_norm + 1e-8)
-            similarities = np.clip(similarities, 0.0, 1.0)
 
             # 应用偏好类型加成
             for nid, sim in zip(valid_node_ids, similarities):
@@ -706,11 +692,7 @@ class PathScoreExpansion:
 
         # 🚀 批量获取记忆对象（如果graph_store支持批量获取）
         # 注意：这里假设逐个获取，如果有批量API可以进一步优化
-        memory_cache: dict[str, Any] = {}
+        memory_cache: dict[str, Any] = self.graph_store.get_memories_by_ids(all_memory_ids)
-        for mem_id in all_memory_ids:
-            memory = self.graph_store.get_memory_by_id(mem_id)
-            if memory:
-                memory_cache[mem_id] = memory
 
         # 构建映射关系
         for path in paths:
@@ -749,30 +731,31 @@ class PathScoreExpansion:
         node_type_cache: dict[str, str | None] = {}
 
         if self.prefer_node_types:
-            # 收集所有需要查询的节点ID
+            # 收集所有需要查询的节点ID，并记录记忆中的类型提示
-            all_node_ids = set()
+            all_node_ids: set[str] = set()
+            node_type_hints: dict[str, str | None] = {}
             for memory, _ in memory_paths.values():
                 memory_nodes = getattr(memory, "nodes", [])
                 for node in memory_nodes:
                     node_id = node.id if hasattr(node, "id") else str(node)
                     all_node_ids.add(node_id)
+                    if node_id not in node_type_hints:
+                        node_obj_type = getattr(node, "node_type", None)
+                        if node_obj_type is not None:
+                            node_type_hints[node_id] = getattr(node_obj_type, "value", str(node_obj_type))
 
-            # 批量获取节点数据
             if all_node_ids:
-                logger.debug(f"🔍 批量预加载 {len(all_node_ids)} 个节点的类型信息")
+                logger.info(f"🧠 预处理 {len(all_node_ids)} 个节点的类型信息")
-                node_data_list = await asyncio.gather(
+                for nid in all_node_ids:
-                    *[self.vector_store.get_node_by_id(nid) for nid in all_node_ids],
+                    node_attrs = self.graph_store.graph.nodes.get(nid, {}) if hasattr(self.graph_store, "graph") else {}
-                    return_exceptions=True
+                    metadata = node_attrs.get("metadata", {}) if isinstance(node_attrs, dict) else {}
-                )
+                    node_type = metadata.get("node_type") or node_attrs.get("node_type")
 
-                # 构建类型缓存
+                    if not node_type:
-                for nid, node_data in zip(all_node_ids, node_data_list):
+                        # 回退到记忆中的节点定义
-                    if isinstance(node_data, Exception) or not node_data or not isinstance(node_data, dict):
+                        node_type = node_type_hints.get(nid)
-                        node_type_cache[nid] = None
-                    else:
-                        metadata = node_data.get("metadata", {})
-                        node_type_cache[nid] = metadata.get("node_type")
 
+                    node_type_cache[nid] = node_type
         # 遍历所有记忆进行评分
         for mem_id, (memory, paths) in memory_paths.items():
             # 1. 聚合路径分数
@@ -868,5 +851,33 @@ class PathScoreExpansion:
 
         return recency_score
 
+    def _batch_compute_similarities(
+        self,
+        valid_embeddings: list["np.ndarray"],
+        query_embedding: "np.ndarray"
+    ) -> "np.ndarray":
+        """
+        批量计算向量相似度（CPU密集型操作，移至to_thread中执行）
+
+        Args:
+            valid_embeddings: 有效的嵌入向量列表
+            query_embedding: 查询向量
+
+        Returns:
+            相似度数组
+        """
+        import numpy as np
+
+        # 批量计算相似度（使用矩阵运算）
+        embeddings_matrix = np.array(valid_embeddings)
+        query_norm = np.linalg.norm(query_embedding)
+        embeddings_norms = np.linalg.norm(embeddings_matrix, axis=1)
+
+        # 向量化计算余弦相似度
+        similarities = np.dot(embeddings_matrix, query_embedding) / (embeddings_norms * query_norm + 1e-8)
+        similarities = np.clip(similarities, 0.0, 1.0)
+
+        return similarities
+
 
 __all__ = ["Path", "PathExpansionConfig", "PathScoreExpansion"]

src/memory_graph/utils/similarity.py

@@ -4,6 +4,7 @@
 提供统一的向量相似度计算函数
 """
 
+import asyncio
 from typing import TYPE_CHECKING
 
 if TYPE_CHECKING:
@@ -47,4 +48,91 @@ def cosine_similarity(vec1: "np.ndarray", vec2: "np.ndarray") -> float:
         return 0.0
 
 
-__all__ = ["cosine_similarity"]
+async def cosine_similarity_async(vec1: "np.ndarray", vec2: "np.ndarray") -> float:
+    """
+    异步计算两个向量的余弦相似度，使用to_thread避免阻塞
+
+    Args:
+        vec1: 第一个向量
+        vec2: 第二个向量
+
+    Returns:
+        余弦相似度 (0.0-1.0)
+    """
+    return await asyncio.to_thread(cosine_similarity, vec1, vec2)
+
+
+def batch_cosine_similarity(vec1: "np.ndarray", vec_list: list["np.ndarray"]) -> list[float]:
+    """
+    批量计算向量相似度
+
+    Args:
+        vec1: 基础向量
+        vec_list: 待比较的向量列表
+
+    Returns:
+        相似度列表
+    """
+    try:
+        import numpy as np
+
+        # 确保是numpy数组
+        if not isinstance(vec1, np.ndarray):
+            vec1 = np.array(vec1)
+
+        # 批量转换为numpy数组
+        vec_list = [np.array(vec) for vec in vec_list]
+
+        # 计算归一化
+        vec1_norm = np.linalg.norm(vec1)
+        if vec1_norm == 0:
+            return [0.0] * len(vec_list)
+
+        # 计算所有向量的归一化
+        vec_norms = np.array([np.linalg.norm(vec) for vec in vec_list])
+
+        # 避免除以0
+        valid_mask = vec_norms != 0
+        similarities = np.zeros(len(vec_list))
+
+        if np.any(valid_mask):
+            # 批量计算点积
+            valid_vecs = np.array(vec_list)[valid_mask]
+            dot_products = np.dot(valid_vecs, vec1)
+
+            # 计算相似度
+            valid_norms = vec_norms[valid_mask]
+            valid_similarities = dot_products / (vec1_norm * valid_norms)
+
+            # 确保在 [0, 1] 范围内
+            valid_similarities = np.clip(valid_similarities, 0.0, 1.0)
+
+            # 填充结果
+            similarities[valid_mask] = valid_similarities
+
+        return similarities.tolist()
+
+    except Exception:
+        return [0.0] * len(vec_list)
+
+
+async def batch_cosine_similarity_async(vec1: "np.ndarray", vec_list: list["np.ndarray"]) -> list[float]:
+    """
+    异步批量计算向量相似度，使用to_thread避免阻塞
+
+    Args:
+        vec1: 基础向量
+        vec_list: 待比较的向量列表
+
+    Returns:
+        相似度列表
+    """
+    return await asyncio.to_thread(batch_cosine_similarity, vec1, vec_list)
+
+
+__all__ = [
+    "cosine_similarity",
+    "cosine_similarity_async",
+    "batch_cosine_similarity",
+    "batch_cosine_similarity_async"
+]