Hacker News new | ask | show | jobs
by zozbot234 104 days ago
That's very large models at full quantization though. Stuff that will crawl even on a decent homelab, despite being largely MoE based and even quantization-aware, hence reducing the amount and size of active parameters.