The Christmas holiday period always brings some opportunity to sneak in some time for hobby projects. Either its because the rest of the family is still sleeping in during the morning or to just sit with the laptop next to the kids while they look a Christmas movie.
During recent years end period there was a lot of excitement about the deepseek model, released open source, so i decided to give it a try to run local.
Unfortunately i don''t have the state of the art hardware anymore which i had as a kid (3df FX Voodoo in SLI) nor didn’t i continue running everything on macbooks and mac mini's as during my student time. Currently all I have running is a not that state of the art windows laptop... so i knew it wouldn’t have good results in token processing time, but that shouldn’t spoil the fun.
First setting up the environment. From my ethical hacker lab I was already used of using oracle box (with Kali), so I continued in that space.
As Kali was for me a bit too specialized, I decided to spin up an Ubuntu vm and followed the tutorial of Pavan:
https://dev.to/pavanbelagatti/run-deepseek-r1-locally-for-free-in-just-3-minutes-1e82
Setting up went fast, although there where as usual some extra libraries to upgrade. Also made the rookie mistake to give the VM insufficient disc space :-D
First impression: slooooooooooooooooooooooooooooooooooow but also what a fun to use!
Didn’t had real experiences with a reasoning model, liked it a lot.
Planned three homelabs as follow-up:
- Using python scripts to call the local LLM (update May: done, see next blog post)
- Play around with different LLMs to measure performance
- Install stable diffusion to create pictures for this blogpost (if hardware allows it...)
SOCIAL SHARE CARD GENERATOR