Generally speaking a home server/workstation set up is going to provide better performance at lower cost. You don't sacrifice much mobility either so long as you have an internet connection and can either SSH tunnel or use Tailscale (never used, just know it's popular).
If Qwen is finetuned for a hardness, it'll be Qwen Code. Qwen 27b works well enough in OpenCode though which is what I use. My one complaint is it likes to get cute with bash commands instead of OpenCode's built-in tools. I use a skill to steer that.
Looping is a common problem with the Qwen models. I've had good luck using --repeat-penalty=1.1 with llama.cpp and 27B. vLLM should have a similar option.
Before you run and go purchase a unified memory computer (e.g., DGX Spark, Mac, Ryzen AI Max 395 / Strix Halo), be aware dense models generally run slow on these machines. Dedicated GPUs run dense models significantly better. Look for benchmarks for your prospective machine. If you really want one of these, you'll be better off running Qwen 3.6 35B or another sparse MoE model.
So I have a library and its ultimate purpose is converting globs to regexes. Someone sent me a ReDoS vulnerability report with a 4.0 CVSS score because if you write an obscene glob pattern you'll get a correspondingly obscene (and inefficient) regex. What else would you have it do!?
You hear the same canard every time Anthropic releases a new model or version. I'm not convinced they're objective anecdotes. I wonder if it's simply the new model, while marginally better, has a different style and people find that new/refreshing. That is what makes it feel so much better than the previous release.
That's the max you can statically allocate in the BIOS. It's best to leave that at the minimum (500 MB I think), and let the drivers dynamically allocate. You can use up to about 120 GB on Linux.
Mac, DGX Spark, and a Framework Desktop / Ryzen AI Max 395 (ie Strix Halo) will not give you great performance running LLMs. One benefit of the Spark over the others is you can easily link up to 4 of them. Only MoE (sparse) models will be usable. Even if you can run some massive models, they will crawl. You're better off running one or more GPU cards.
Even if houses were half the cost, buying and selling one would be a long, cumbersome process that I would not want to go through just to live in a city short term. You're running off the false assumption that no one wants to rent.
> for housing we need to replace zoning rules with a neighbor vote, put the community back in control instead of bureaucrats who block new apartments for dumb reasons or take bribes from developers to build luxury investment properties nobody wants.
> there should also probably be an escape hatch where anyone can build affordable/rent controlled apts without a vote but only if the location has high enough average rent.
A microwave is orders of magnitude cheaper and easier to purchase than housing. In fact, they appear to be too cheap to be available at Aaron's for rent. So you're going to have to cough up $50 for a microwave from Walmart if you only want one for two months.
The landlords could then pay rents to the local government. We could call that landlord taxes or something and use that to fund local services with them. Wait...