Shipped her own ashxhart 's TensorFold kit for GLM-5.3-Flash on two DGX Sparks.
I ran the full agent grid on it last night. The recipe's own numbers check out, and the context story is the real news 2 DGX Sparks and Mac M5 Ultra. THE SETUP Same pair of Sparks,
You asked how to run ego (lite) with a local model.
So we pulled Qwen 3.8 with Ollama, hooked it up to OpenCode, and pointed it at Airbnb Tokyo. A CSV with listings, prices, and ratings comes back. No API key. No cloud.
I'm not a dev nor an engineer guy.
Just an enthusiast with some compute. It has the overthinking flaw, but it made the best pagoda in my pc to the day 💪 Im now downloading the swift 1.5 for less thinking You should test it. Is fast and it runs at the same speed as qwen 3.8 27b
Strata qwen 3.8 flash on a 3090 card and I'll add a Pic of sys specs.
Local llm. Code focus. Only like 35gig used included is the OS for the machine, Ubuntu. So this can run while I am running over things for sure, easily. Aug is around 75 to 85tps tokens per sec. Strata is a
You have reached the end of the archive
All of qwen38