New in town
Get started: use the models on your own machine
Never run an AI model yourself before? This is the whole trail, start to finish. Download a model from the corral, install one small app, and you will be talking to it on your own computer in about ten minutes — no keys, no bills, nothing leaving your machine.
Four steps, start to finish
- 01
Create your free account
Signing in lets you save models to your outfit, follow updates, and post in the Saloon when you get stuck. Browsing and downloading stay free.
- 02
Pick a model that fits your machine
On a model page, check the size. A 7–8B model at 4-bit needs about 6–8 GB of memory. 13–14B wants 16 GB. 3B-class models run on most laptops without a graphics card.
- 03
Download the files
Every ungated model page has direct download links for its weight files. Click the primary file to pull it straight to your computer, or copy the command line shown on the page.
- 04
Run it with a local app
You need one small program that loads the file and gives you a chat box or an API. Pick one of the three paths below — Ollama is the easiest place to start.
Three ways to run it
Choose one. They all load the same downloaded files.
Ollama — easiest
Best first stop. One install, one command, works on Mac, Windows, and Linux.
# 1. Install from ollama.com, then:
ollama run qwen2.5:7b
# chat right in the terminal, or call it from code:
curl http://localhost:11434/api/generate -d '{"model":"qwen2.5:7b","prompt":"Howdy"}'LM Studio — no terminal
A desktop app with a chat window. Point it at a downloaded GGUF file and start typing.
1. Install LM Studio 2. Open the folder icon and load your .gguf file 3. Chat in the app, or flip on the local server 4. Your app can then call http://localhost:1234/v1
Python — most control
For builders who want the model inside their own code or a high-throughput server.
pip install transformers torch
from transformers import pipeline
pipe = pipeline("text-generation", model="./qwen-7b")
print(pipe("Howdy partner,", max_new_tokens=64)[0]["generated_text"])Going deeper
Once the first model is running, size up bigger ones, quantize to fit tighter memory, and serve them behind a local endpoint.
Questions from the trail
Do I need a graphics card to use these models?
No. Smaller models (roughly 3B and under, or 4-bit builds of 7B) run on a normal laptop CPU, just slower. A GPU or an Apple Silicon Mac makes bigger models comfortable.
What is a GGUF file and which file should I download?
GGUF is a single-file format built for local apps like Ollama and LM Studio — that is usually the one you want. Folders of .safetensors files are for Python and GPU serving.
Is anything sent to the internet when I run a model?
Only the initial download. After that the model runs on your machine, so your prompts and documents never leave it.
Do I have to pay to download models?
No. Weights on ModelCorral are free to download. Check each model's license for how you may use it commercially.
Where do I get help if a model will not load?
Ask in the Saloon. Post your machine, the model name, and the error, and the community — or the Sheriff — will get you moving.