I found out a $40 open source model runs on my old laptop with 8GB of RAM and I did not believe the blog post at first
I was reading a blog post last Tuesday from a guy in Bangalore who showed a quantized 7B model running on a five year old ThinkPad with no GPU at all, just 8GB of RAM and a custom loader. I tried it myself the next night and the thing answered my test questions in about 3 seconds each, which blew my mind compared to the cloud tools I was paying $20 a month for. Is anyone here running local models on old hardware, and if so what setups have worked for you?