#VibeThinker-1.5B
1 messages · Page 1 of 1 (latest)
Come on now lol
LMo
Their paper doesn't seem vibed, at least
Parent company seems fairly relevant, wonder what this is all about
hmmm 🤔
slop
only 1.5B is tiny
obviously benchmaxxed :p
It's definitely benchmaxxed, they do state it's the goal (math and reasoning heavy questions)
I can't decide if I'm impressed that they managed to do it to such a small model
so its MoE but they just cut away all Experts other than one
(jk)
If its actually really good in the domain it was trained for then I'd be impressed
This reminded me of https://huggingface.co/Zyphra/ZR1-1.5B as they are both based on Qwen2.5-Math-1.5B
Part of me likes it. They’re making a good point about these benchmarks.
Agreed we need more of that