Bet. Looking into that now. Thanks!
I believe I have 11g of vram, so I should be good to run decent models from what I’ve been told by the other AIs.
Comment on Frustratingly bad at self hosting. Can someone help me access LLMs on my rig from my phone
brucethemoose@lemmy.world 20 hours ago
At risk of getting more technical, ik_llama.cpp has a fantastic built in webui:
github.com/ikawrakow/ik_llama.cpp/
Getting more technical, its also way better than ollama. You can run models way smarter than ollama can on the same hardware.
For reference, I’m running GLM-4 (667 GB of raw weights) on a single RTX 3090/Ryzen gaming rig, at reading speed, with pretty low quantization distortion.
And if you want a ‘look this up on the internet for me’ assistant (which you need for them to be truly useful), you need another docker project as well.
…That’s just how LLM self hosting is now. It’s simply too hardware intense to be easy. You can indeed host a small LLM without much understanding, but its going to be pretty dumb.
BlackSnack@lemmy.zip 20 hours ago
brucethemoose@lemmy.world 19 hours ago
In case I miss your reply, assuming a 3080 + 64 GB of RAM, you want the IQ4_KSS (or IQ3_KS, for more RAM for tabs and stuff) version of this:
huggingface.co/ubergarm/GLM-4.5-Air-GGUF
Part of it will run on your GPU, part will live in system RAM, but ik_llama.cpp does the quantizations split in a particularly efficient way for these kind of ‘MoE’ models.
If you ‘only’ have 32GB RAM or less, that’s tricker, and the next question is what kind of speeds do you want. But it’s probably best to wait a few days and see how Qwen3 80B looks when it comes out. Or just go with the IQ4_K version of this: huggingface.co/…/Qwen3-30B-A3B-Thinking-2507-GGUF
And you don’t really need the hyper optimization of ik_llama.cpp for Qwen3 30B.
Alternatively, you could try to squeeze Gemma 27B into that 11GB VRAM, but it would be tight.
brucethemoose@lemmy.world 19 hours ago
How much system RAM, and what kind? DDR5?
tal@lemmy.today 19 hours ago
Ollama does have some features that make it easier to use for a first-time user, including:
Calculating automatically how many layers can fit in VRAM and loading that many layers and splitting between CPU and VRAM. kobold.cpp can’t do that automatically yet.
Automatically unloading the model from VRAM after a period of inactivity.
I had an easier time setting up ollama than other stuff, and OP does apparently already have it set up.
brucethemoose@lemmy.world 19 hours ago
Yeah. But it also messes stuff up from the llama.cpp baseline, and hides or doesn’t support some features/optimizations, and definitely doesn’t support the more efficient iq_k quants of ik_llama.cpp and its specialzied MoE offloading.
And that’s not even getting into the various controversies around ollama (like broken GGUFs or indications they’re going closed source in some form).
…It just depends on how much performance you want to squeeze out, and how much time you want to spend on the endeavor. Small LLMs are kinda marginal though, so its pretty important IMO.