As I now have a machine that has enough RAM to run some serious local models (see Framework Desktop), I ran some preliminary experiments using a codebase that Tyler Shoemaker and I have been working on to test the limits of open source LLMs such as the qwen family of models. (I have been trying to find hardware to run these on for a while, and thought I had SSH access to a machine with enough RAM at one stage a few weeks ago: but then it turned out something was off with my login credentials, or possibly the VPN I use to access the network.) More next week on this front, I hope.