I use a 4B model (gemma 4) as a compression model, to handle larger and local SWE workflows with context <100k on systems, like Strix halo on my Hermes setup.
Right? Wasn’t there that story locally where this guy had an AI email digest of stock price movements but he realized the AI was hallucinating like months later?
DiscipleofDeceit666@reddit
What do you use 4b for?
rbm1@reddit
I use a 4B model (gemma 4) as a compression model, to handle larger and local SWE workflows with context <100k on systems, like Strix halo on my Hermes setup.
AnticitizenPrime@reddit
Gemma4 4B might be the best lightweight vision model there is, too.
ab2377@reddit (OP)
in opencode, web searches, i read news on it. run docker, write scripts, little good and nice and fast things.
DiscipleofDeceit666@reddit
But it hallucinates tho, like a lot right?
ab2377@reddit (OP)
not at all. don't give it things that 32b models should be doing.
DiscipleofDeceit666@reddit
Like what? I never trusted those models bc when I used them for code related things (like stack trace analysis) it was pretty garbage
spawncampinitiated@reddit
And he just said he uses it to web fetch news
DiscipleofDeceit666@reddit
Right? Wasn’t there that story locally where this guy had an AI email digest of stock price movements but he realized the AI was hallucinating like months later?
MaxKruse96@reddit
qwhen
Abject-Kitchen3198@reddit
First word on my mind
Evening_Ad6637@reddit
You should care also about good qwentizations
fustercluck6000@reddit
In a little qwhile
some_user_2021@reddit
qwhile(!qwen) qomplain();
kevinlch@reddit
qwhen.gguf when?
some_user_2021@reddit
Qwhen will then be now?
ab2377@reddit (OP)
i like this word a lot!
craftogrammer@reddit
qwomorrow
HitarthSurana@reddit
greedy qoblins
grabber4321@reddit
you probably didnt even use 3.6 properly yet.
Iory1998@reddit
There is no Qwen3.6-4B lol
ab2377@reddit (OP)
and i loved it https://www.reddit.com/r/LocalLLaMA/s/V9VFuVwh0U
ab2377@reddit (OP)
i did what my gpu poor allowed.
dondiegorivera@reddit
Someone on Linkedin calculated tomorrow based on recent Alibaba release patterns. Let’s hope the best.
FoxiPanda@reddit
Somewhere between now and never.
KillaDan365@reddit
I ain't gonna live forever
ab2377@reddit (OP)
pretty sure they will release it tomorrow
Comacdo@reddit
Why so sure ?
KURD_1_STAN@reddit
He just picked the middle between now and never, s simple decision really nothing to be sure about
ab2377@reddit (OP)
i had a dream!
OMG_IM_A_GIRL@reddit
Calm down Ramanujan.
No_Swimming6548@reddit
Roughly in 20 days, or never
info_solutions@reddit
Ask Alibaba Cloud !
freehuntx@reddit
qwen slowly drifts to cloud only. Sorry buddy.
EveningIncrease7579@reddit
ab2377@reddit (OP)
in opencode, web searches, i read news on it. run docker, write scripts, little good and nice and fast things.