Skip to content

Personal AI, On Personal Devices

Python

fix: skip torch-dependent tests when torch not installed in CI

Jon Saad-Falcon258feff64e78
diff --git a/tests/connectors/test_embedding_store.py b/tests/connectors/test_embedding_store.py
index d63c49e4..002a424a 100644
--- a/tests/connectors/test_embedding_store.py
+++ b/tests/connectors/test_embedding_store.py
@@ -9,6 +9,8 @@ import pytest
from openjarvis.connectors.embedding_store import EmbeddingStore
+torch = pytest.importorskip("torch", reason="torch required for embedding tests")
+
# ---------------------------------------------------------------------------
# Fixtures
# ---------------------------------------------------------------------------
diff --git a/tests/connectors/test_retriever.py b/tests/connectors/test_retriever.py
index 80d011c5..453055a8 100644
--- a/tests/connectors/test_retriever.py
+++ b/tests/connectors/test_retriever.py
@@ -12,6 +12,15 @@ from openjarvis.connectors.retriever import ColBERTReranker, Reranker, TwoStageR
from openjarvis.connectors.store import KnowledgeStore
from openjarvis.tools.storage._stubs import RetrievalResult
+
+def _has_torch() -> bool:
+ try:
+ import torch # noqa: F401
+ return True
+ except ImportError:
+ return False
+
+
# ---------------------------------------------------------------------------
# Fixtures
# ---------------------------------------------------------------------------
@@ -215,6 +224,9 @@ def test_retrieve_recall_k_larger_than_top_k(
# ---------------------------------------------------------------------------
+ not _has_torch(), reason="torch required for embedding tests",
+)
def test_reranker_uses_cached_embeddings() -> None:
"""ColBERTReranker checks EmbeddingStore.get() before calling docFromText().
@@ -265,6 +277,9 @@ def test_reranker_uses_cached_embeddings() -> None:
# ---------------------------------------------------------------------------
+ not _has_torch(), reason="torch required for embedding tests",
+)
def test_reranker_caches_new_embeddings() -> None:
"""When EmbeddingStore.get() returns None, docFromText() is called and
the result is stored back via EmbeddingStore.store()."""