I needed to use the GPU today, so I couldn't use the local llm all day.
I don't know how many times a day I check if an official version supporting MTP has been released.
So, I decided to install the beta version of llama.cpp that supports MTP.
Since I don't know much about installation, I asked hermes agent to do it. I used qwen3.6-plus for the brain (it's my most satisfying model personally. It's smart and not too expensive...). It seems like I spent about a dollar.
When I got stuck, AI당 provided information, told me the GitHub address, and even where to find GGUF, so it was easy to figure out.
When I need to use the GPU, I have to stop the local llm, so I asked them to make a skill for that too, and they made it right away.
If you say "turn off the llama server" in the prompt, it will turn it off, and if you say "turn it on," it will turn it on.
I haven't been able to experience MTP yet because I still need to use the GPU, but I wanted to brag about it on AI당.
I'm really curious about its performance... I thought about getting a censorship-removed model, but I was too scared... haha
My system is Ryzen 5700x / DDR4 96GB / RTX3090FE.
