Post

aly
@aly.codes

yeah no this model fucks A lot of the sub 30B models are pretend-useful but 30B+ has some real sweet spots for local

aly@aly.codes

qwen 3.8 flash next is surprisingly capable for how much VRAM you need to run it

aly
@aly.codes

I’m using strata, getting very good performance, and can simultaneously load up a 30B dense model alongside it. (I think muse glimmer is the most usable 30B ish model, just wish it was MoE)

0 reposts18 likes

No replies yet.