Meta Llama 3.1 is Game Over for GPT 4o ❓

Meta AI released Llama 3.1 family of models that include the biggest 405B, a model containing 405 billion parameters. Llama 3.1 405B is GPT-4 class model and scores 88.6 on MMLU benchmark, only behind GPT-4o which is at 88.7 MMLU.

Trained using 16,000 Nvidia H100 GPUs, it also benefits from newer training and development techniques that Meta claims makes it competitive with leading proprietary models like OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet.

In a paper published today, researchers at the company write that they’re actively developing Llama models that can recognize images and videos, and understand (and generate) speech. Still, these models aren’t yet ready for public release.

#ai #meta #llama3.1

[Analyticsvidhya]

Book FREE 1:1 Mentorship for Gen AI / Data Science

[jamad-ym]

Zuck must have watched Gladiator too many times growing up

[zachio]

Zuck looks like he finally took his quest headset off

[mduzondi]

"you don't have to send your data to another company" I didn't think that will ever come out of Mark's own mouth. Mark is data and data is Mark

[wilddropbear]

Yeah but could you kindly make it more open source by removing all Censorship Please.

[CRIMELAB]

Idc if u a reptilian, thank you for llama

[tangoolo]

Open source model will be a thing like Linux ONLY when the computing power, infra, needed to train these models becomes generally "feasible". 400B param means, roughly, 400GBs of VRAM, which is very very.. expensive for any normal geek in a basement.

[endremoen]

Some documentation on system req for the smaler 70b and 8b models would be good. Aswell as precission like 4, 8, 16 together with layer pruning for the most common gpus. Eg Im trying to install 70b on rtx3090 with 32 gb of ddr4 memory

[meraculus-znzx]

I understand why they call him a lizard person now

[pablov.viteri]

Where do I rent a good machine to run it?

[the_agz]

As time passes Lama users can show off like arch users does