Repository navigation
[Bug] Some of think models are returning incorrect display formats. #140
Description
Activity
@Jeffrey-deng could you please copy the RAW message and paste it here? Look for the "Copy" button under the LLM reply when you hover over it.
As shown previously, there is only a single closing tag.
Okay, so I'm trying to figure out how to respond to the user's "hello world" message. From the initial prompt, it looks like the user wants me to act as an experienced software engineer who follows a specific process when reviewing and suggesting code changes. First, I need to understand the structure they expect. They mentioned using <CODE_REVIEW>, <PLANNING>, and then providing the correct output. But in this case, the user just said "hello world," which is a simple program, not a code review scenario. Hmm, maybe the user is testing how I handle different types of inputs. Since "hello world" is a common first program, perhaps they want me to explain it or suggest improvements. But according to the guidelines, I should conduct a code review first. However, there's no existing code to review here. I should probably ask for clarification. Maybe the user wants to discuss the "hello world" program in more depth, like different implementations or best practices. Or perhaps they have a specific problem related to it that they haven't mentioned yet. I don't want to assume too much, so it's better to ask. That way, I can provide a more accurate and helpful response. I'll structure my response to acknowledge their message and request more details about what they need help with regarding "hello world." </think> It seems like you're introducing a "hello world" program, which is a common first step in learning programming. However, I don't see any existing code to review. Could you clarify what you'd like to discuss or achieve with this program? Are you looking for an explanation, different implementations, or perhaps best practices? Let me know how I can assist you!I found the API documentation for siliconflow:
https://docs.siliconflow.cn/cn/userguide/capabilities/reasoning
ENGLISH:
https://docs.siliconflow.cn/en/userguide/capabilities/reasoningIf it is a reasoning model, the "thinking" part returns in the reasoning_content field.
data: {"id":"019b4b3633d030b2726a1631bed875fa","object":"chat.completion.chunk","created":1766493402,"model":"deepseek-ai/DeepSeek-R1","choices":[{"index":0,"delta":{"content":null,"reasoning_content":"首先","role":"assistant"},"finish_reason":null}],"system_fingerprint":"","usage":{"prompt_tokens":10,"completion_tokens":1,"total_tokens":11,"completion_tokens_details":{"reasoning_tokens":1}}}data: {"id":"019b4b3633d030b2726a1631bed875fa","object":"chat.completion.chunk","created":1766493402,"model":"deepseek-ai/DeepSeek-R1","choices":[{"index":0,"delta":{"content":"整理方程","reasoning_content":null,"role":"assistant"},"finish_reason":null}],"system_fingerprint":"","usage":{"prompt_tokens":10,"completion_tokens":1213,"total_tokens":1223,"completion_tokens_details":{"reasoning_tokens":839}}}@Jeffrey-deng at this moment,
reasoning_contentis not handled for Completions API, and the tag seems to be actual model output/glitch.
I'll keep this ticket open to consider reasoning support in Completions API
Describe the bug
The thought process is displayed in the main text and ends with a tag.
I am using the SiliconFlow DeepSeek-R1-Distill-Qwen-32B model.
To Reproduce
Steps to reproduce the behavior:
Expected behavior
A clear and concise description of what you expected to happen.
Screenshots
If applicable, add screenshots to help explain your problem.
Environment:
Additional context
Add any other context about the problem here (for example, exported file with chats)