Is it actually possible? Well...

...yes, but not really. The closest you're going to get is a service like Azure or AWS. The actual hardware for running Deepseek V3 is incredibly expensive. You need around 500 gigabytes of RAM (slower) or VRAM (faster), which will set you back somewhere in the area of $2500-$250000. HOWEVER, you can run a "distilled" model.

What are those?

Distilled models are AI models released by other companies, that were trained by Deepseek to increase performance. They are much smaller, and have less performance, but can run on most consumer hardware. There are many versions to fit each size, including on 8GB or less of RAM, although they are worse that the full model (although not by too much!).

Well, guide me, magician!

Follow these OS-specific instructions, then read to the end. The links will directly go to ollama.com, which is the way we are going to run our Deepseek distills.

Windows

Windows Installer (ollama)

Linux

curl -fsSL https://ollama.com/install.sh | sh
Or, Manually Install

MacOS 11+

MacOS Installer (ollama)

Continuing...

I recommend restarting your machine to stop any issues, specifically looking at you, windows users (it didn't work for me otherwise).

Open your terminal, and run this command:

ollama run deepseek-r1

This command runs the 7b version of deepseek-r1, which should run on most tech. If you want a larger model, add a ":32b" to the end of that "deepseek-r1", or ":1.5b" if you need to go smaller. After it finishes downloading, you should just have the model available for use at any time. Obviously, your speed and maximum size allowed will vary depending on your RAM and VRAM (RAM but for graphics card, quicker). If you have over 40GB of VRAM and/or RAM spare, you can, try ":70b", and you can try the full model at "671b". Have fun with whatever you do from this!


PS, you may want to look into https://github.com/open-webui/open-webui if this seems cool to you, it runs on docker.