hm, q8 qwen distilled is 21gb but works fine without multigpu. I have 16gb vram. I expected to need the node with virtual vram (gguf multigpu).

Q8 distilled sample, 29s with the 8 step lora