The agent engineering platform.
Python agentsaiai-agentsanthropicchatgptdeepagentsenterpriseframeworkgeminigenerative-ailangchainlanggraphllmmultiagentopen-sourceopenaipydanticpythonragtypescript
Fixes: community: fix LanceDB return no metadata (#27024)
binhnd102f723a8456eba
diff --git a/libs/community/langchain_community/vectorstores/lancedb.py b/libs/community/langchain_community/vectorstores/lancedb.pyindex 7253d885b1..f08e438048 100644--- a/libs/community/langchain_community/vectorstores/lancedb.py+++ b/libs/community/langchain_community/vectorstores/lancedb.py@@ -151,12 +151,14 @@ class LanceDB(VectorStore):score_col = "_relevance_score"else:score_col = None+ # Check if 'metadata' is in the columns+ has_metadata = "metadata" in columnsif score_col is None or not score:return [Document(page_content=results[self._text_key][idx].as_py(),- metadata=results["metadata"][idx].as_py(),+ metadata=results["metadata"][idx].as_py() if has_metadata else {},)for idx in range(len(results))]@@ -165,7 +167,9 @@ class LanceDB(VectorStore):(Document(page_content=results[self._text_key][idx].as_py(),- metadata=results["metadata"][idx].as_py(),+ metadata=results["metadata"][idx].as_py()+ if has_metadata+ else {},),results[score_col][idx].as_py(),)diff --git a/libs/community/tests/integration_tests/vectorstores/test_lancedb.py b/libs/community/tests/integration_tests/vectorstores/test_lancedb.pyindex 7ba3a00466..615b310629 100644--- a/libs/community/tests/integration_tests/vectorstores/test_lancedb.py+++ b/libs/community/tests/integration_tests/vectorstores/test_lancedb.py@@ -114,3 +114,41 @@ def test_lancedb_all_searches() -> None:)assert len(result_3[0]) == 2 # type: ignoreassert "text 1" in result_3[0][0].page_content # type: ignore++[email protected]("lancedb")+def test_lancedb_no_metadata() -> None:+ lancedb = import_lancedb()+ embeddings = FakeEmbeddings()+ # Connect to a temporary LanceDB instance+ db = lancedb.connect("/tmp/lancedb_no_metadata_test")+ # Create data without the 'metadata' field+ texts = ["text 1", "text 2", "item 3"]+ data = []+ for idx, text in enumerate(texts):+ embedding = embeddings.embed_documents([text])[0]+ data.append(+ {+ "vector": embedding,+ "id": str(idx),+ "text": text,+ # Note: We're deliberately not including 'metadata' here+ }+ )+ # Create the table without 'metadata' column+ db.create_table("vectorstore_no_metadata", data=data)+ # Initialize LanceDB with the existing connection and table name+ store = LanceDB(+ connection=db,+ embedding=embeddings,+ table_name="vectorstore_no_metadata",+ )+ # Perform a similarity search+ result = store.similarity_search("text 1")+ # Verify that the metadata in the Document objects is an empty dictionary+ for doc in result:+ assert (+ doc.metadata == {}+ ), "Expected empty metadata when 'metadata' column is missing"+ # Clean up by deleting the table (optional)+ db.drop_table("vectorstore_no_metadata")