Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

fix(embedder): 稀疏嵌入仅提交文本输入 (#4296)

Axiomoth41044af77901
diff --git a/openviking/models/embedder/volcengine_embedders.py b/openviking/models/embedder/volcengine_embedders.py
index 1f6d74d82..bbec5e4ef 100644
--- a/openviking/models/embedder/volcengine_embedders.py
+++ b/openviking/models/embedder/volcengine_embedders.py
@@ -405,8 +405,16 @@ class VolcengineSparseEmbedder(SparseEmbedderBase):
@property
def supports_multimodal(self) -> bool:
- """Multimodal inputs are supported when using the multimodal endpoint."""
- return self.input_type == "multimodal"
+ """Sparse vectors require text-only input even on the multimodal endpoint.
+
+ The provider accepts the request at the endpoint level, but rejects
+ non-text parts (images and videos) whenever ``sparse_embedding`` is
+ enabled. Returning ``False`` makes the common embedder guard retain the
+ extracted text and drop those parts before the provider call. Image-only
+ search remains unsupported and is rejected by the retrieval capability
+ check instead of embedding empty text.
+ """
+ return False
class VolcengineHybridEmbedder(HybridEmbedderBase):
@@ -489,8 +497,14 @@ class VolcengineHybridEmbedder(HybridEmbedderBase):
@property
def supports_multimodal(self) -> bool:
- """Hybrid embeddings always use the multimodal endpoint."""
- return True
+ """Hybrid vectors include sparse output and therefore require text input.
+
+ A hybrid provider request enables sparse embedding, for which image and
+ video parts are unsupported. The base guard performs a deterministic
+ text-only downgrade for mixed content. Image-only search remains
+ unsupported because dropping its only input would produce an empty query.
+ """
+ return False
def embed(self, content: "EmbeddingInput", is_query: bool = False) -> EmbedResult:
"""Perform hybrid embedding on text or multimodal content
diff --git a/tests/unit/test_volcengine_embedder_input.py b/tests/unit/test_volcengine_embedder_input.py
new file mode 100644
index 000000000..e0f74b19d
--- /dev/null
+++ b/tests/unit/test_volcengine_embedder_input.py
@@ -0,0 +1,32 @@
+# Copyright (c) 2026 Beijing Volcano Engine Technology Co., Ltd.
+# SPDX-License-Identifier: AGPL-3.0
+"""Input-guard tests for Volcengine sparse-capable embedding modes."""
+
+from unittest.mock import patch
+
+from openviking.models.embedder.volcengine_embedders import (
+ VolcengineHybridEmbedder,
+ VolcengineSparseEmbedder,
+)
+
+_MULTIMODAL_INPUT = [
+ {"type": "text", "text": "diagram of a heat exchanger"},
+ {"type": "image_url", "image_url": {"url": "https://example.test/image.png"}},
+ {"type": "video_url", "video_url": {"url": "https://example.test/video.mp4"}},
+]
+
+
+@patch("openviking.models.embedder.volcengine_embedders.volcenginesdkarkruntime.Ark")
+def test_sparse_embedder_downgrades_multimodal_input_to_text(_mock_ark):
+ embedder = VolcengineSparseEmbedder(model_name="test", api_key="test-key")
+
+ assert embedder.supports_multimodal is False
+ assert embedder.prepare_embedding_input(_MULTIMODAL_INPUT) == "diagram of a heat exchanger"
+
+
+@patch("openviking.models.embedder.volcengine_embedders.volcenginesdkarkruntime.Ark")
+def test_hybrid_embedder_downgrades_multimodal_input_to_text(_mock_ark):
+ embedder = VolcengineHybridEmbedder(model_name="test", api_key="test-key")
+
+ assert embedder.supports_multimodal is False
+ assert embedder.prepare_embedding_input(_MULTIMODAL_INPUT) == "diagram of a heat exchanger"