zhudongwork

I tried update #14065 , but it still didn't work.

> @zhudongwork Thanks for your issue. Could you please upload a pptx file with the problems. And we can quickly locate the bug. [demo.pptx](https://github.com/user-attachments/files/19547564/demo.pptx)

> Thank you for submitting the code. You have resolved the error issues we had before, but the effect of the version you generated in the specific text parsing is...

The splitter method of Dify is not correct. Generally, the separator is used on the red part, but Dify is used to split large documents.

But when the input is long(~60k), the agent node returns an empty output after running for a few seconds. Is there a timeout setting causing this? With the same input,...

When is this feature expected to go online? I noticed that the dify-plugin has added the `create_retriever_resource_message` method, but dify itself does not yet support this message type.

Another kind of parallel stream output exception of the Question Classifier node. https://github.com/langgenius/dify/issues/15700

> 我相信代币``应该``基于[教程](https://github.com/LLaVA-VL/LLaVA-NeXT/blob/b3a46be22d5aa818fa1a23542ae3a28f8e2ed421/docs/LLaVA_OneVision_Tutorials.ipynb)（在`Video Input`部分） > > 我这样做了，并且能够正常微调（使用 LoRA） 如果您明白这一点，请告诉我：在推理时（合并后），我得到了乱码输出。 作为参考，我的基础模型是`lmms-lab/llava-onevision-qwen2-0.5b-ov` Have you encountered a situation where after fine-tuning a LoRA model and loading it for inference, the responses are confusing?

In the official environment, after upgrading the agent policy plugin to the latest version, I still cannot achieve the agent node reference display functionality.

失效了，再来一个