262·
4 days ago“Open source” in ML is a really bad description for what it is. “Free binary with a bit of metadata” would be more accurate. The code used to create deepseek is not open source, nor is the training datasets. 99% of “open source” models are this way. The only interesting part of the open sourcing is the architecture used to run the models, as it lends a lot of insight into the training process, and allows for derivatives via post-training
When I started learning Linux at work, the game I played with myself was i’d install Debian stable minimal on my primary workstation and I would not reinstall it ever. No matter what happened, I would always fix it.
I learned to install the basic subsystems to get a GUI and audio, learned the fun of Nvidia drivers to get xinerama and hw decoding working. In retrospect it seems trivial but as a new learner it was challenging and rewarding.
At one point I was trying to do something, and a guide online suggested installing some repo and installing newer libraries. I did so, and a week later I did a dist-upgrade (because I didn’t know any better) and when I rebooted I was presented with a splash screen for “crunchbang” linux.
Figuring out how to get back to Debian without breaking everything probably taught me more about packages, package managers, filesystems, system config files, init (systemd wasn’t really a thing yet) than everything else I had done combined.
For anyone wondering: 12 years into the project I had a drive from the mdadm mirror die, and while mdadm was copying to another mirror, the other drive died. I considered that a win but y’all can be the judge (no files were lost, 12yr into my Linux journey I had long since figured out automating NFS and rsync).