perf: reuse embedding server for query embed

packages/leann-core/src/leann/interface.py

@@ -2,13 +2,16 @@ from abc import ABC, abstractmethod
 import numpy as np
 from typing import Dict, Any, List, Literal
 
+
 class LeannBackendBuilderInterface(ABC):
     """Backend interface for building indexes"""
-    
-    @abstractmethod 
-    def build(self, data: np.ndarray, ids: List[str], index_path: str, **kwargs) -> None:
+
+    @abstractmethod
+    def build(
+        self, data: np.ndarray, ids: List[str], index_path: str, **kwargs
+    ) -> None:
         """Build index
-        
+
         Args:
             data: Vector data (N, D)
             ids: List of string IDs for each vector
@@ -17,30 +20,35 @@ class LeannBackendBuilderInterface(ABC):
         """
         pass
 
+
 class LeannBackendSearcherInterface(ABC):
     """Backend interface for searching"""
-    
+
     @abstractmethod
     def __init__(self, index_path: str, **kwargs):
         """Initialize searcher
-        
+
         Args:
             index_path: Path to index file
             **kwargs: Backend-specific loading parameters
         """
         pass
-    
+
     @abstractmethod
-    def search(self, query: np.ndarray, top_k: int,
-               complexity: int = 64,
-               beam_width: int = 1,
-               prune_ratio: float = 0.0,
-               recompute_embeddings: bool = False,
-               pruning_strategy: Literal["global", "local", "proportional"] = "global",
-               zmq_port: int = 5557,
-               **kwargs) -> Dict[str, Any]:
+    def search(
+        self,
+        query: np.ndarray,
+        top_k: int,
+        complexity: int = 64,
+        beam_width: int = 1,
+        prune_ratio: float = 0.0,
+        recompute_embeddings: bool = False,
+        pruning_strategy: Literal["global", "local", "proportional"] = "global",
+        zmq_port: int = 5557,
+        **kwargs,
+    ) -> Dict[str, Any]:
         """Search for nearest neighbors
-        
+
         Args:
             query: Query vectors (B, D) where B is batch size, D is dimension
             top_k: Number of nearest neighbors to return
@@ -51,23 +59,40 @@ class LeannBackendSearcherInterface(ABC):
             pruning_strategy: PQ candidate selection strategy - "global" (default), "local", or "proportional"
             zmq_port: ZMQ port for embedding server communication
             **kwargs: Backend-specific parameters
-            
+
         Returns:
             {"labels": [...], "distances": [...]}
         """
         pass
 
+    @abstractmethod
+    def compute_query_embedding(
+        self, query: str, zmq_port: int = 5557, use_server_if_available: bool = True
+    ) -> np.ndarray:
+        """Compute embedding for a query string
+
+        Args:
+            query: The query string to embed
+            zmq_port: ZMQ port for embedding server
+            use_server_if_available: Whether to try using embedding server first
+
+        Returns:
+            Query embedding as numpy array with shape (1, D)
+        """
+        pass
+
+
 class LeannBackendFactoryInterface(ABC):
     """Backend factory interface"""
-    
+
     @staticmethod
     @abstractmethod
     def builder(**kwargs) -> LeannBackendBuilderInterface:
         """Create Builder instance"""
         pass
-    
+
     @staticmethod
-    @abstractmethod  
+    @abstractmethod
     def searcher(index_path: str, **kwargs) -> LeannBackendSearcherInterface:
         """Create Searcher instance"""
-        pass
+        pass