Hunting for a personal agent brain on sub-12GB hardware
The author says 1-bit quants in the sub-12GB range are only good for chatbots and is downloading a Qwen 27B distillation to keep testing models for a local personal agent.
I'm yet to see a 1-bit quant (for the sub 12GB range) worth its salt for anything more than a chatbot, and 0xSero's previous 1-bit recommendations (like original Bonsai 1-bit prior) were underwhelming. But.. I appreciate the updated listings, and I'll never turn down an interesting new Qwen 27B distillation for testing. My search for the ultimate personal agent brain (on sub 12GB hardware) continues... Underdog Saluki 27B 1.0, now downloading...
This post quotes another post: “Local model picks by VRAM tier, from 4GB to 384GB”