diff --git a/src/memos/memories/textual/general.py b/src/memos/memories/textual/general.py index b90f2a6ab..6bfa45597 100644 --- a/src/memos/memories/textual/general.py +++ b/src/memos/memories/textual/general.py @@ -128,9 +128,7 @@ def search(self, query: str, top_k: int, info=None, **kwargs) -> list[TextualMem """ query_vector = self._embed_one_sentence(query) search_results = self.vector_db.search(query_vector, top_k) - search_results = sorted( # make higher score first - search_results, key=lambda x: x.score, reverse=True - ) + # Keep the backend's relevance order: distance and similarity scores rank differently. result_memories = [ TextualMemoryItem(**search_item.payload) for search_item in search_results ] diff --git a/tests/memories/textual/test_general.py b/tests/memories/textual/test_general.py index bebedcb56..0f5e1f38a 100644 --- a/tests/memories/textual/test_general.py +++ b/tests/memories/textual/test_general.py @@ -4,6 +4,8 @@ from unittest.mock import MagicMock, patch +import pytest + from memos.configs.embedder import EmbedderConfigFactory from memos.configs.llm import LLMConfigFactory from memos.configs.memory import GeneralTextMemoryConfig @@ -16,6 +18,45 @@ from memos.vec_dbs.item import VecDBItem +@pytest.mark.parametrize("distance_metric", ["euclidean", "cosine", "dot"]) +@pytest.mark.parametrize("top_k", [2, 3]) +def test_search_preserves_qdrant_relevance_order(tmp_path, distance_metric, top_k): + config = GeneralTextMemoryConfig( + extractor_llm={"backend": "ollama", "config": {"model_name_or_path": "test-llm"}}, + embedder={"backend": "ollama", "config": {"model_name_or_path": "test-embedder"}}, + vector_db={ + "backend": "qdrant", + "config": { + "collection_name": "ranking_test", + "vector_dimension": 2, + "distance_metric": distance_metric, + "path": str(tmp_path / "qdrant"), + }, + }, + ) + embeddings = { + "query": [1.0, 0.0], + "near": [1.0, 0.0], + "middle": [0.5, 0.5], + "far": [-1.0, 0.0], + } + embedder = MagicMock() + embedder.embed.side_effect = lambda texts: [embeddings[text] for text in texts] + with ( + patch("memos.memories.textual.general.LLMFactory.from_config"), + patch("memos.memories.textual.general.EmbedderFactory.from_config", return_value=embedder), + ): + memory = GeneralTextMemory(config) + + try: + memory.add([TextualMemoryItem(memory=text) for text in ["middle", "far", "near"]]) + assert [item.memory for item in memory.search("query", top_k)] == ["near", "middle", "far"][ + :top_k + ] + finally: + memory.vector_db.client.close() + + class TestGeneralTextMemory(unittest.TestCase): def setUp(self): # Mock configurations for GeneralTextMemoryConfig arguments