diff --git a/reme2/component/embedding/base_embedding_model.py b/reme2/component/embedding/base_embedding_model.py
index c3493978..6bf73b99 100644
--- a/reme2/component/embedding/base_embedding_model.py
+++ b/reme2/component/embedding/base_embedding_model.py
@@ -28,7 +28,7 @@ class BaseEmbeddingModel(BaseComponent):
         pass_dimensions: bool = False,
         max_batch_size: int = 10,
         max_input_length: int = 8192,
-        max_cache_size: int = 5000,
+        max_cache_size: int = 10000,
         enable_cache: bool = True,
         max_retries: int = 3,
         **kwargs,
@@ -99,7 +99,7 @@ class BaseEmbeddingModel(BaseComponent):
                 for orig_idx, text, emb in zip(indices, texts, embeddings):
                     if emb is None:
                         continue
-                    emb_array = np.asarray(emb, dtype=np.float32)
+                    emb_array = np.asarray(emb, dtype=np.float16)
                     if len(emb_array) != self.dimensions:
                         if len(emb_array) < self.dimensions:
                             emb_array = np.pad(emb_array, (0, self.dimensions - len(emb_array)))
@@ -172,7 +172,7 @@ class BaseEmbeddingModel(BaseComponent):
                 continue
             if len(self._embedding_cache) >= self.max_cache_size:
                 break
-            self._embedding_cache[str(key)] = emb.astype(np.float32)
+            self._embedding_cache[str(key)] = emb.astype(np.float16)
 
     def _save_cache(self) -> None:
         if not self.enable_cache or not self._embedding_cache:
diff --git a/reme2/component/file_store/local_file_store.py b/reme2/component/file_store/local_file_store.py
index 9292e850..f523bc52 100644
--- a/reme2/component/file_store/local_file_store.py
+++ b/reme2/component/file_store/local_file_store.py
@@ -133,12 +133,12 @@ class LocalFileStore(BaseFileStore):
         if not query_embedding:
             return []
 
-        candidates = [c for c in self.file_chunks.values() if c.embedding]
+        candidates = [c for c in self.file_chunks.values() if c.embedding is not None]
         if not candidates:
             return []
 
-        candidate_embeddings = np.array([c.embedding for c in candidates])
-        similarities = batch_cosine_similarity(np.array([query_embedding]), candidate_embeddings)[0]
+        candidate_embeddings = np.stack([c.embedding for c in candidates])
+        similarities = batch_cosine_similarity(query_embedding.reshape(1, -1), candidate_embeddings)[0]
 
         results = [
             c.model_copy(update={"scores": {"vector": float(s), "score": float(s)}})
diff --git a/reme2/schema/emb_node.py b/reme2/schema/emb_node.py
index 36011d23..7a261d3a 100644
--- a/reme2/schema/emb_node.py
+++ b/reme2/schema/emb_node.py
@@ -1,10 +1,26 @@
 from uuid import uuid4
 
-from pydantic import BaseModel, Field
+import numpy as np
+from pydantic import BaseModel, Field, field_serializer, field_validator, ConfigDict
 
 
 class EmbNode(BaseModel):
+    model_config = ConfigDict(arbitrary_types_allowed=True)
+
     id: str = Field(default_factory=lambda: uuid4().hex)
     text: str = Field(default="")
-    embedding: list[float] | None = Field(default=None)
+    embedding: np.ndarray | None = Field(default=None)
     metadata: dict = Field(default_factory=dict)
+
+    @field_validator('embedding', mode='before')
+    @classmethod
+    def validate_embedding(cls, v):
+        if v is None:
+            return v
+        return np.array(v, dtype=np.float16)
+
+    @field_serializer('embedding')
+    def serialize_embedding(self, v: np.ndarray | None, _info):
+        if v is None:
+            return None
+        return v.tolist()
diff --git a/reme2/schema/file_chunk.py b/reme2/schema/file_chunk.py
index 4f8077cc..b30d803a 100644
--- a/reme2/schema/file_chunk.py
+++ b/reme2/schema/file_chunk.py
@@ -16,4 +16,4 @@ class FileChunk(EmbNode):
     def set_hash_id(self):
         from ..utils import hash_text
         self.id = hash_text(" ".join([self.path, str(self.start_line), str(self.end_line), self.text]))
-        return self.id
\ No newline at end of file
+        return self
\ No newline at end of file