Qwen released QwQ-32B - a reasoning model with performance comparable to DeepSeek-R1 on many benchmarks. However, people have been experiencing infinite generations, many repetitions, <think> token issues and finetuning issues. I hope this guide will help debug and fix most issues!

I hope this amount of effort means multi-GPU support is coming soon (saw you mention “this week” :) ) so we can tune QwQ on a single node!

Yes!! Technically it's already in Unsloth :) I'm still trying to make the frontend for running multi GPU better!