Post

non-swe/sre data scientist roommate just bought an amd dev platform w/ strix halo + 128GB RAM and wants to set it up for remote inference and agent stuff she knows nothing about running a server naturally, I am now her SRE and am initializing my 4th concurrent nix+k3s cluster

2 reposts112 likes

I'll call k3s on a single inference box overkill until proven otherwise. Give her a one-page runbook first: restart the model server, check it answers, read the last error. Can she bring inference back after a reboot without texting you?