[Bug]: Input sequence length exceeds the max input length of embedding model.

[lichangqing2611]

Bug Description

我们采用OV 0.2.6，embedding模型采用bce-embedding-base_v1-f16.gguf，这个embeddding模型max输入长度是512，physical batch size是1024，当openclaw对话结束后，还会持续几次embedding调度，但这几次调用的输入tokens会最长到几K，如8129，会报下面的错

Steps to Reproduce

Bug Info:

2026-03-18 11:02:50,993 - openviking.storage.queuefs.semantic_processor - INFO - Completed semantic generation for: viking://user/man/memories/preferences
2026-03-18 11:02:52,458 - openviking.storage.collection_schemas - ERROR - Error processing embedding message: OpenAI API error: Error code: 500 - {'error': {'code': 500, 'message': 'input (8129 tokens) is too large to process. increase the physical batch size (current batch size: 1024)', 'type': 'server_error'}}
Traceback (most recent call last):
  File "/usr/local/lib64/python3.11/site-packages/openviking/models/embedder/openai_embedders.py", line 99, in embed
    response = self.client.embeddings.create(**kwargs)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/openai/resources/embeddings.py", line 132, in create
    return self._post(
           ^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/openai/_base_client.py", line 1294, in post
    return cast(ResponseT, self.request(cast_to, opts, stream=stream, stream_cls=stream_cls))
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/openai/_base_client.py", line 1067, in request
    raise self._make_status_error_from_response(err.response) from None
openai.InternalServerError: Error code: 500 - {'error': {'code': 500, 'message': 'input (8129 tokens) is too large to process. increase the physical batch size (current batch size: 1024)', 'type': 'server_error'}}

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/usr/local/lib64/python3.11/site-packages/openviking/storage/collection_schemas.py", line 197, in on_dequeue
    result: EmbedResult = await asyncio.to_thread(
                          ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.11/asyncio/threads.py", line 25, in to_thread
    return await loop.run_in_executor(None, func_call)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.11/concurrent/futures/thread.py", line 58, in run
    result = self.fn(*self.args, **self.kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib64/python3.11/site-packages/openviking/models/embedder/openai_embedders.py", line 104, in embed
    raise RuntimeError(f"OpenAI API error: {e.message}") from e
RuntimeError: OpenAI API error: Error code: 500 - {'error': {'code': 500, 'message': 'input (8129 tokens) is too large to process. increase the physical batch size (current batch size: 1024)', 'type': 'server_error'}}

Expected Behavior

No Error.

Actual Behavior

ERROR

Minimal Reproducible Example

Error Logs

OpenViking Version

0.2.6

Python Version

3.11.0

Operating System

Linux

Model Backend

Other

Additional Context

No response

[wu-xiaochen]

This is a common issue when dealing with long-context embeddings. For ontology-based systems like OpenViking, consider implementing hierarchical context chunking - splitting large contexts into smaller manageable pieces before embedding. You might also want to look into embedding models that support longer context windows (e.g., bge-large-zh-v1.5 supports 512, but newer models like text-embedding-3 support up to 8k). Implementing a sliding window approach for memory consolidation could help prevent these overflow issues.

[qin-ctx]

感谢反馈！这个问题已收录到统一跟踪 issue #744 中。

分类：Embedding 处理与分块策略

当前 embedding 流程缺乏对模型 max_input_length 的适配，导致超长输入直接报错。我们正在优化：

• 在 vectorization 前增加长度检测
• 针对不同 embedding 模型自动适配 max_tokens
• 引入智能分块策略，超长内容自动 chunk 后分别 embed

请关注 #744 获取最新进展。