Hacker News
DaisyRose-27B
woah
|next
[-]
anon373839
|root
|parent
|next
[-]
harry8
|root
|parent
|next
[-]
0xWTF
|root
|parent
|previous
[-]
its_down_again
|root
|parent
|next
|previous
[-]
charcircuit
|root
|parent
|previous
[-]
It already happened. https://github.com/karpathy/autoresearch
jacquesm
|next
|previous
[-]
nblintao
|root
|parent
|next
[-]
jacquesm
|root
|parent
[-]
mik1998
|root
|parent
|next
[-]
jacquesm
|root
|parent
[-]
reissbaker
|root
|parent
|previous
[-]
jacquesm
|root
|parent
[-]
And no, that won't get you to AGI, that is a completely different goal.
banghua13
|root
|parent
[-]
jacquesm
|root
|parent
[-]
pkoiralap
|next
|previous
[-]
https://huggingface.co/sabicap/sabi-noema
Also HF model here: https://huggingface.co/RadixArk/DaisyRose-27B
xlayn
|next
|previous
[-]
"18 days on 144 H200 GPUs" insane.. didn't thought so much hardware would be needed to train Qwen.
drdaeman
|root
|parent
|next
[-]
cookiengineer
|root
|parent
|previous
[-]
Also this is on consumer hardware, meaning 96GB CPU RAM and 48GB GPU VRAM, using an optimized Vulkan concurrent shader pipeline. So costs for the machine are around 3-5k depending on RAM crisis.
mips_avatar
|next
|previous
[-]
0x69420
|next
|previous
[-]
robertkarl
|next
|previous
[-]
octoberfranklin
|next
|previous
[-]
You might be able to shave that down using a different flavor of H200 (like PCIe) but it would take more hours, so benefits are unclear.
mips_avatar
|root
|parent
|next
[-]
lumost
|root
|parent
|previous
[-]
octoberfranklin
|root
|parent
[-]
But seriously. If there were a cheaper and reliable way, I'm sure the authors would have mentioned it instead of H200 SXM modules. They're trying to brag about cheap training.
Maybe you wanted to use an opportunity to shill tranium.