Interesting. I wonder if model distillation affected censoring in R1.

I think we might be talking about separate things. I tested with this 32B distilled model using llama-cpp.

Just to clarify - DeepSeek censors its hosted service. Self-hosted models aren’t affected.

NixOS does suck me into the occasional rabbit hole - nothing like Gentoo years ago though!

I don’t miss XML, but at least it has support for comments. On the other hand, I wish whitespace in XML wasn’t significant. JSON needs to die in favor of JSON 5.