Skip to content

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

fix(embedding): truncate plain list inputs to max_input_tokens (#5760) (#5771)

akushonkamen54e199f41df3
diff --git a/openviking/models/embedder/openai_embedders.py b/openviking/models/embedder/openai_embedders.py
index a0a76e9e8..909061bec 100644
--- a/openviking/models/embedder/openai_embedders.py
+++ b/openviking/models/embedder/openai_embedders.py
@@ -2,7 +2,7 @@
# SPDX-License-Identifier: AGPL-3.0
"""OpenAI Embedder Implementation"""
-from typing import Any, Dict, List, Literal, Optional
+from typing import Any, Dict, List, Literal, Optional, cast
import httpx
import openai
@@ -22,6 +22,7 @@ from openviking.models.network import (
from openviking.models.vlm.registry import DEFAULT_AZURE_API_VERSION
from openviking.telemetry import get_current_telemetry
from openviking.utils.async_client_cache import LoopScopedAsyncClientCache
+from openviking.utils.embedding_input import truncate_embedding_input
from openviking_cli.utils import get_logger
logger = get_logger(__name__)
@@ -322,7 +323,10 @@ class OpenAIDenseEmbedder(DenseEmbedderBase):
def _prepare_embedding_input(self, text: EmbeddingInput | List[str]) -> EmbeddingInput | List[str]:
if isinstance(text, list) and all(isinstance(item, str) for item in text):
- return text
+ if self.max_input_tokens is None:
+ return text
+ text = cast(List[str], text)
+ return [truncate_embedding_input(item, self.max_input_tokens) for item in text]
return self.prepare_embedding_input(text)
def _build_kwargs(
diff --git a/tests/unit/test_openai_embedder.py b/tests/unit/test_openai_embedder.py
index 5ca3090ef..e812ce0f5 100644
--- a/tests/unit/test_openai_embedder.py
+++ b/tests/unit/test_openai_embedder.py
@@ -87,7 +87,7 @@ class TestOpenAIDenseEmbedder:
@patch("openviking.models.embedder.openai_embedders.openai.OpenAI")
def test_embed_preserves_openai_batch_text_input(self, mock_openai_class):
- """OpenAI batch text input should keep the historical list[str] passthrough."""
+ """OpenAI batches keep item boundaries and truncate each oversized item."""
mock_client = MagicMock()
mock_openai_class.return_value = mock_client
@@ -101,13 +101,13 @@ class TestOpenAIDenseEmbedder:
embedder = OpenAIDenseEmbedder(
model_name="text-embedding-3-small",
api_key="test-api-key",
- dimension=1536,
+ config={"max_input_tokens": 1},
)
- embedder.embed(["first text", "second text"])
+ embedder.embed(["first text", "x"])
call_kwargs = mock_client.embeddings.create.call_args[1]
- assert call_kwargs["input"] == ["first text", "second text"]
+ assert call_kwargs["input"] == ["firs\n...(truncated for embedding)", "x"]
@patch("openviking.models.embedder.openai_embedders.openai.OpenAI")
def test_multimodal_input_type_passes_content_parts(self, mock_openai_class):