Posts

@BennettBuhner ohh ok so they likely reprioritised

i can see why it’s good! ppl who have proper GPUs/rigs they can run these on def benefit so that’s nice

@BennettBuhner bc also iirc sam said he wanted to have the open source model be able to run on phones, and i don’t think many phones will be able to comfortably run an 8b model

@BennettBuhner when i say idle, i mean with apps that i usually have open (a browser, discord, etc..)

@BennettBuhner i think 8b ones would be slow since macOS already uses 5-7GB of RAM idle – when i use the smaller gemma/qwen models locally i always go for 1-3b param models, 4b+ is pushing it

@BennettBuhner The leaks look really cool but I’m worried about the smallest OS model being 20B since I only have 8GB of ram on my Mac :P

Ig if it’s open source someone could make a 1b or 4b param model like Gemma