this post was submitted on 13 Feb 2025
7 points (88.9% liked)

LocalLLaMA

3230 readers
1 users here now

Welcome to LocalLLaMA! Here we discuss running and developing machine learning models at home. Lets explore cutting edge open source neural network technology together.

Get support from the community! Ask questions, share prompts, discuss benchmarks, get hyped at the latest and greatest model releases! Enjoy talking about our awesome hobby.

As ambassadors of the self-hosting machine learning community, we strive to support each other and share our enthusiasm in a positive constructive way.

Rules:

Rule 1 - No harassment or personal character attacks of community members. I.E no namecalling, no generalizing entire groups of people that make up our community, no baseless personal insults.

Rule 2 - No comparing artificial intelligence/machine learning models to cryptocurrency. I.E no comparing the usefulness of models to that of NFTs, no comparing the resource usage required to train a model is anything close to maintaining a blockchain/ mining for crypto, no implying its just a fad/bubble that will leave people with nothing of value when it burst.

Rule 3 - No comparing artificial intelligence/machine learning to simple text prediction algorithms. I.E statements such as "llms are basically just simple text predictions like what your phone keyboard autocorrect uses, and they're still using the same algorithms since <over 10 years ago>.

Rule 4 - No implying that models are devoid of purpose or potential for enriching peoples lives.

founded 2 years ago
MODERATORS
 

I have an GTX 1660 Super (6 GB)

Right now I have ollama with:

  • deepseek-r1:8b
  • qwen2.5-coder:7b

Do you recommend any other local models to play with my GPU?

top 6 comments
sorted by: hot top controversial new old
[–] [email protected] 3 points 4 months ago* (last edited 4 months ago)

I have a 1070ti 6gb so right there with you. Its important to note though that our use cases and expectations may differ. Also I'm using kobold.cpp with cublas partial offloading to run the models

Qwen 14B R1 distill q6km for testing out CoT, science/math related questions, internet search RAG, and best token speed to performance ratio

Arliai Mistral NeMo 12B finetune q4km for smut and creative writing.

Beepo 22b Mistral Small 2407 uncensored fine tune model that will tell me all the forbidden no-no knowledge.

Mistral Small 3 2501 for the best generally performing model that can fit on the card with bearable token speed and context window.

Minicpm for multimodal vision for document scanning.

[–] [email protected] 2 points 4 months ago (1 children)

Deepseek is good at reasoning, qwen is good at programming, but I find llama3.1 8b to be well suited for creativity, writing, translations and other tasks which fall out of the scope of your two models. It's a decent all arounder. It's about 4.9GB in q4_K_M.

[–] [email protected] 1 points 4 months ago

It's not out of my scope, I'm just learning what can I do locally with my current machine.


Today I read about RAG, maybe I'm gonna try an easy local setup to chat with a PDF.

[–] [email protected] 2 points 4 months ago (1 children)

Mistral

I personally run models on my laptop. I have 48 GB of ram and a i5-12500U. It runs a little slow but usable

[–] [email protected] 2 points 4 months ago (1 children)

My gear is an old:

I7-4790 16GB RAM

How many tokens by second?

[–] [email protected] 1 points 4 months ago

The biggest bottleneck is going to be memory. I would just stick with GPU only since your GPU memory has the most bandwidth.