fix(data): clarify _nothink suffix warning for reasoning-only models#10613
Conversation
The previous warning unconditionally advised adding `_nothink` suffix, which is misleading for model families like Qwen3.6 that only ship reasoning variants and have no Instruct counterpart. Closes hiyouga#10566 Co-Authored-By: Claude <noreply@anthropic.com>
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates the warning message in src/llamafactory/data/template.py to clarify when the _nothink suffix should be used with reasoning templates. The reviewer suggested refining the terminology in the warning message—specifically replacing "base model" with "model" and clarifying the parenthetical explanation—to avoid confusing users regarding base versus instruct models.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
| "You are using reasoning template. " | ||
| "If the base model is NOT a reasoning model (i.e., it has a separate Instruct variant), " | ||
| "please add `_nothink` suffix to disable thinking. " | ||
| "For reasoning-only model families (e.g., Qwen3.6), the suffix is not needed. " | ||
| "e.g., qwen3_vl_nothink" |
There was a problem hiding this comment.
在 LLM 社区中，“base model”（基座模型）通常特指未经指令微调（SFT）的预训练模型，而“Instruct variant”则是指对话/指令模型。在这里使用“base model”可能会让用户产生混淆（例如，用户可能会认为自己使用的是 Instruct 模型，因此不属于 base model）。
此外，“it has a separate Instruct variant” 的表述也有些令人困惑，因为如果用户正在使用非推理的 Instruct 模型，该模型本身就是 Instruct 版本，而不是“拥有”一个独立的 Instruct 版本。
建议将 “base model” 改为 “model”，并将括号内的说明调整为更准确的表述，例如 (e.g., a standard Instruct variant)。
| "You are using reasoning template. " | |
| "If the base model is NOT a reasoning model (i.e., it has a separate Instruct variant), " | |
| "please add `_nothink` suffix to disable thinking. " | |
| "For reasoning-only model families (e.g., Qwen3.6), the suffix is not needed. " | |
| "e.g., qwen3_vl_nothink" | |
| "You are using reasoning template. " | |
| "If the model is NOT a reasoning model (e.g., a standard Instruct variant), " | |
| "please add '_nothink' suffix to disable thinking. " | |
| "For reasoning-only model families (e.g., Qwen3.6), the suffix is not needed. " | |
| "e.g., qwen3_vl_nothink" |
2 participants
Fixes #10566
get_template_and_fix_tokenizer中的警告无条件建议添加_nothink后缀，对于仅发布推理模型（无 Instruct 版本）的模型系列（如 Qwen3.6）会产生误导。修改警告文本，明确说明：
This contribution was prepared with AI assistance.