What Ollama models do you use? Is there a good repository of rankings?
So I imagine this community has a bunch folks on the Ollama train, and I admit that I have a bit of choice paralysis in this. I've tried a half dozen 8b-14b models, but would love to get some kind of consensus on things. I am aware of hugging face and the like, but I'll openly admit I don't know how most of these benchmarks work, and haven't found a really solid "Plain English ranking" for selfhosted models. I know that https://openwebui.com/leaderboard Open Webui's leaderboard exists, but it goes too far the other direction, why are you at the top?
TL;DR: What models are you using at home? Ideally one that can be run on consumer cards. Love to know if you're using it cause its faster, or if its the most accurate, or both?
Open WebUI
LLM Leaderboard - Open WebUI
LLM leaderboard from the Open WebUI community - help us create the best community leaderboard by sharing your feedback history!
2 Replies
I kinda would love this to be more a discussion, apologies if its not the right place? But I can get us started.
I've been using the Qwen2.5 8 and 14b models, and their coder versions. I also just downloaded the Microsoft phi4, and it seems noticably faster than the Qwen, and may have actually better answer in my inane tests.
GPT4chan-8B never steered me wrong for coding
jokes aside I really like phi4 or QwQ 32B
I run it on my 4090