Skip to content

利用 AI 大模型和自动化工作流,根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.

fix(twelvelabs): reject mixed embedding dimensions

Rudy Celekli01af233fc2bf
diff --git a/app/services/twelvelabs.py b/app/services/twelvelabs.py
index aa32b629..a4d1328b 100644
--- a/app/services/twelvelabs.py
+++ b/app/services/twelvelabs.py
@@ -135,8 +135,10 @@ def rerank_terms_by_subject(
scored = []
for term in search_terms:
vec = embed_text(term, model)
- if vec is None:
- # If any term can't be embedded, don't risk a partial reorder.
+ if vec is None or len(vec) != len(subject_vec):
+ # Cosine requires vectors from the same dimensional space. zip()
+ # would silently truncate a malformed response and produce a
+ # misleading score, so retain the all-or-nothing fallback.
return search_terms
scored.append((term, _cosine(subject_vec, vec)))
diff --git a/test/services/test_twelvelabs_embedding_dimensions.py b/test/services/test_twelvelabs_embedding_dimensions.py
new file mode 100644
index 00000000..848cbfd3
--- /dev/null
+++ b/test/services/test_twelvelabs_embedding_dimensions.py
@@ -0,0 +1,35 @@
+from contextlib import contextmanager
+from types import SimpleNamespace
+
+import pytest
+
+from app.services import twelvelabs as service
+
+
[email protected]("invalid_vector", [[1.0], [1.0, 0.0, 0.0]])
+def test_rerank_preserves_original_order_for_mixed_dimensions(monkeypatch, invalid_vector):
+ monkeypatch.setitem(service.config.app, "twelvelabs_api_keys", ["synthetic"])
+ monkeypatch.setitem(service.config.app, "twelvelabs_rerank_terms", True)
+ vectors = {"subject": [1.0, 0.0], "orthogonal": [0.0, 1.0], "malformed": invalid_vector}
+
+ @contextmanager
+ def managed_client():
+ yield SimpleNamespace(embed=SimpleNamespace(create=lambda **kw: SimpleNamespace(
+ text_embedding=SimpleNamespace(segments=[SimpleNamespace(float_=vectors[kw["text"]])])
+ )))
+
+ monkeypatch.setattr(service, "_managed_client", managed_client)
+ service._embed_text_cached.cache_clear()
+ try:
+ terms = ["orthogonal", "malformed"]
+ assert service.rerank_terms_by_subject("subject", terms) is terms
+ finally:
+ service._embed_text_cached.cache_clear()
+
+
+def test_matching_dimensions_still_rank_by_similarity(monkeypatch):
+ monkeypatch.setattr(service, "is_enabled", lambda: True)
+ monkeypatch.setitem(service.config.app, "twelvelabs_rerank_terms", True)
+ vectors = {"subject": [1.0, 0.0], "orthogonal": [0.0, 1.0], "aligned": [1.0, 0.0]}
+ monkeypatch.setattr(service, "embed_text", lambda text, model=None: vectors[text])
+ assert service.rerank_terms_by_subject("subject", ["orthogonal", "aligned"]) == ["aligned", "orthogonal"]