nshotton

you could try using a local vision model, like Mage-VL from microsoft. Its only a 5b model so its quite small for the capability it has.

it looks like it says the 27b will be included in the release too, we will find out in 30 minutes lol

you can see the countdown here: https://modelscope.cn/models/Qwen/Qwen3.8-2.4T-A95B?spm=a2ty...

Thanks for posting this, it made a huge difference tweaking the recipe.

I might be missing something here, but could this concept not also be used in the same way for harm? For example if a model is trained to replicated a sucessfull scammer?

flash is 284b isnt it? https://artificialanalysis.ai/models/deepseek-v4-flash

Yeah i shouldve been more clear, a model of this size could run on 2 dgx sparks so out of the range of a lot of the typical consumer sure, but I think there is definitely a market for that size

This model is shockingly small for how capable it is. its a little bit bigger than deepseekV4 flash but around as capable if not more on some benchmarks than V4 pro, i wouldnt be surprised if this becomes a popular local model.

sweet, is there any plans to implement VLLM in the future?

This looks really great! Quick question, is there any support for using local models?

I'm struggling to see how this could lower the cost to ideate here, almost all of the designs I can see on your page would require re drafting/designing it would be more beneficial for people to look at homes that have already been built compared to an AI generation with no real world grounding. Also who is paying for the compute to generate these floor plans and renders? This would have to turn into a paid service eventually im assuming.