Haven't tried, would be surprised if it's any different.

It's new arch demo for future Qwen 4 family, but (as I understand) training recipe/data is same as any other 3.8 model.