src/pyseekdb/utils/embedding_functions/__init__.py (1)

38-40: Import placement and __all__ ordering inconsistency.

The import is placed after __all__, unlike all other imports which are at the top. Additionally, the __all__ list appears to follow alphabetical order, but Text2VecEmbeddingFunction is appended at the end instead of between TencentHunyuanEmbeddingFunction and VoyageaiEmbeddingFunction.

♻️ Suggested fix

Move the import to the top with other imports:

 from .tencent_hunyuan_embedding_function import TencentHunyuanEmbeddingFunction
+from .text2vec_embedding_function import Text2VecEmbeddingFunction
 from .voyageai_embedding_function import VoyageaiEmbeddingFunction

And update __all__ ordering:

     "TencentHunyuanEmbeddingFunction",
+    "Text2VecEmbeddingFunction",
     "VoyageaiEmbeddingFunction",
-    "Text2VecEmbeddingFunction",
 ]
-from .text2vec_embedding_function import Text2VecEmbeddingFunction

tests/unit_tests/test_text2vec_embedding.py (1)

98-110: Silence unused variable warnings with _ prefix.

The ef and ef2 variables are assigned to trigger model loading side effects but are never read. Prefix with _ to indicate intentional non-use and silence the linter warnings (F841).

♻️ Suggested fix

         # First init should load model
-        ef = Text2VecEmbeddingFunction(model_name="new-model")
+        _ef = Text2VecEmbeddingFunction(model_name="new-model")
         mock_sentence_model.assert_called_once()

         # Second init with same model should not call constructor again
-        ef2 = Text2VecEmbeddingFunction(model_name="new-model")
+        _ef2 = Text2VecEmbeddingFunction(model_name="new-model")
         mock_sentence_model.assert_called_once()  # Call count remains 1

src/pyseekdb/utils/embedding_functions/text2vec_embedding_function.py (2)

64-72: Cache the dimension property to avoid repeated model inference.

Every access to dimension triggers an encode() call, which is expensive. Consider caching the result after the first computation.

♻️ Suggested fix

+    _cached_dimension: int | None = None
+
     `@property`
     def dimension(self) -> int:
         """Get the dimension of embeddings produced by this function."""
+        if self._cached_dimension is not None:
+            return self._cached_dimension
         # Get dimension from the model's encoding directly if possible
         # Or try encoding a dummy string
         sample = self._model.encode("test", normalize_embeddings=self.normalize_embeddings)
         if hasattr(sample, 'shape'):
-            return sample.shape[0] if len(sample.shape) == 1 else sample.shape[1]
-        return len(sample)
+            self._cached_dimension = sample.shape[0] if len(sample.shape) == 1 else sample.shape[1]
+        else:
+            self._cached_dimension = len(sample)
+        return self._cached_dimension

---

35-38: Consider recursive validation for nested structures (optional).

The validation accepts list, dict, and tuple but doesn't recursively validate their contents. A nested structure containing a non-serializable object (e.g., {"key": some_callable}) would pass validation but fail serialization. This is likely acceptable for typical usage, but worth noting.