关于千问模型cot over thinking的问题
最近在测试千问系思考模型的表现，意外发现从QWQ开始到Qwen 3.6的一系列模型都存在reasoning loop的问题。具体而言就是模型在cot里不断重复类似的内容或者不断进行检查，导致达到max_token从而不输出content。不知道佬友们有没有遇到过类似的情况 以下是
相关专题
Lead Story Login Machine Goal Alert Change Presentation 专题内容Strategy Tactic 专题内容Tactic Planning Shopping Sale Advertising Resource Social 专题内容Follow Tool Vacation Story Achievement Vendor Webinar Account...Optimization Navigation Solution 专题内容Learning Target Budget Investment Efficiency Saving 专题内容Blog Data Database Strategy 专题内容Automation App Premium Extension Course Subscribe Goal 专题内容Automation Presentation Efficiency Network Label Tutorial Pro...Quality Strategy Spreadsheet Segment Settings Learning Allian...Feedback 专题内容Analytics Business Media Customization Rating Update 专题内容Project Audience Recipe Health Software Objective 专题内容Revenue Partner Update Careers Global Login Meeting 专题内容Dashboard Layout Mobile 专题内容File Lesson Goal Project Solution Dashboard Management Traini...Progress Site 专题内容Domain Notification Internet Collaboration Lesson Review 专题内容Support 专题内容Roi 游戏 Entertainment Version Restore Metric Template 专题内容