bartowski

/

QwQ-32B-Preview-abliterated-GGUF

like 22

Open the model settings and change the prompt template to ChatML.

I found that mixing GPU offload values results in gibberish also, so you can try offloading 100% to the GPU.