Is there some KDL chart for MiMo-V2.5 or something regarding the quants quality?

I'm using the model with opencode and the issue is it's looping hard when reasoning. It's not a deranged babbling though, the reasoning is legit large spans of text, it just can't get outside of the loop and make a decision. So I babysit it, stop and direct it to the right path, and it mostly does o

中文全文翻译正在补齐，可先查看下方英文原文。

I'm using the model with opencode and the issue is it's looping hard when reasoning. It's not a deranged babbling though, the reasoning is legit large spans of text, it just can't get outside of the loop and make a decision. So I babysit it, stop and direct it to the right path, and it mostly does o