Comment by petu

Unsloth one is gguf for llama.cpp (and some other on-device engines).

So advantage is not having to produce your own quantisation / gguf from .safetensors you've linked.