fix: cached retriever model weights

retrieval/models/sentence_bert.py

@@ -9,14 +9,20 @@ from sentence_transformers import SentenceTransformer
 
 from ..base import BaseRetriever
 
+_model_cache: Dict[str, SentenceTransformer] = {}
+
 
 class DenseRetriever(BaseRetriever):
     """Dense retrieval using sentence transformers."""
-    
+
     def __init__(self, model_name: str, db_path: str, index_config: Literal["naive", "v1"]):
         super().__init__(db_path, index_config)
         self.model_name = model_name
-        self.model = SentenceTransformer(model_name)
+        if model_name not in _model_cache:
+            model = SentenceTransformer(model_name)
+            model.half()  # fp16: halves model memory footprint
+            _model_cache[model_name] = model
+        self.model = _model_cache[model_name]
         self.corpus_embeddings = None
         self.embeddings_path = self._default_embeddings_path()
         self.load_index() if self._embeddings_exist() else self.build_index()