[Question]:Ollama version 0.12.5 removed the "think" content from the content field. How can I modify the code to meet the requirement of displaying the reasoning process?

[husl616]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (Language Policy).
• Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• Please do not modify this template :) and fill in all the required fields.

Describe your problem

Starting from Ollama version 0.9.0, the "think" content is no longer included in the content field but has been moved to a separate think field. However, I still want to include the "think" content within the content field to ensure compatibility with RAGFlow's existing UI, which displays the reasoning process. I'm using RAGFlow v0.20.04. In the case of streaming output, which parts of RAGFlow's code should I modify? thanks.

[dosubot]

Hi @husl616! I'm Dosu and I’m helping the ragflow team.

To merge the "think" content from Ollama 0.9.0+ back into the main "content" field for streaming output in RAGFlow v0.20.04, you should modify the streaming response logic in the backend. The key place to do this is inside the stream_output method of the Generate class (usually found in agent/component/generate.py). This method yields each chunk of the LLM response to the UI, and you can intercept and modify the content before it is sent.

If your LLM response object contains both content and think fields, you can concatenate them (for example, by prepending or appending the think value to content, optionally wrapping it in <think> tags for UI compatibility). Do this inside the loop that yields each streaming chunk. Here’s a simplified example of where to add your logic:

for ans in chat_mdl.chat_streamly(...):
    # Suppose ans is a dict with 'content' and 'think'
    merged_content = ans.get("content", "")
    think_content = ans.get("think", "")
    if think_content:
        merged_content = f"<think>{think_content}</think>\n{merged_content}"
    res = {"content": merged_content, "reference": []}
    yield res

If your streaming response is not a dict, you may need to adjust how you parse the Ollama output upstream (for example, in the Ollama adapter or LLM wrapper) so that both fields are accessible in each chunk.

RAGFlow’s UI expects reasoning steps to be inside the content field, ideally wrapped in <think> tags for proper rendering. There’s no built-in workaround for this; it must be handled in your backend code before yielding to the UI. For more details and examples, see this thread and this explanation of UI rendering requirements.

If you run into edge cases (like missing think fields or different model output formats), you may need to add conditional logic to handle those gracefully.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}