Meta claims ‘world’s largest' open AI model with Llama 3.1 405B debut

Meta today released Llama 3.1 405B, its largest and most capable large language model yet, which the social network claims can go toe-to-toe with OpenAI and Anthropic's top models. "Our experimental evaluation suggests that our flagship model is competitive with leading foundation models across a range of tasks, including GTP- …

[beast666]

I'll bet it's a big bag o shite.

[Anonymous Coward]

no. no it's not

in fact, in my humble it is one of the best.

[Irongut]

Re: no. no it's not

Did it teach you to write without capitalisation, dropping important words and making no sense?

In my opinion, your opinion is worthless because you can't even write a very simple sentence correctly.

[Sorry that handle is already taken.]

Re: no. no it's not

To be fair, it can be both "one of the best" and "a big bag o shite" simultaneously

[Anonymous Coward]

Re: no. no it's not

u so clever with ur big words. I am so impressed that you countered my argument by criticising my form. Very poor and a sign of a low IQ. :poking-out-tongue-returd-style-emoji

As to why the comment of mine was so brief and deliberately omitted the word opinion from the phrase 'in my humble' - most with an IQ above 70 would understand the register of the sentence being crafted intentionally that way.

I can use big letters for people like you who are hard of thinking but it comes across as SHOUTING.

My opinion, boy, is worth X,XXX,XXX in money coins. And as much as we all hate Facebook and Mark, it is just pure stupidity to not recognise that the man is talented (a scum bag as well) he played a blinder fooling Apple with the Meta stuff whilst working tirelessly on Llama.

In fact, I saw a video of him talking about Trump shooting and he looked ... cool. Big chain around his neck and dressed in hip-hop trainers and smart gear. Moved away from that super casual scruffy look at last which has plagued our business. Never thought I would think such a thing, but there you are. Life.

Meta doesn't really get much press except here on the Reg.

And to you, IronGut (name says it all ... we are not dealing with an intellectual here, but instead a person driven by primeval instincts), what are you? Nothing but my b1tch now.

[John70]

If it used Facebook's data for learning it'll be more than just a bag o shite.

[a pressbutton]

more than just a bag o shite

In fb i see many adverts from dropshippers for things that may well be a bag of shite.

I welcome FBs advances in advertising accuracy.

[Sorry that handle is already taken.]

A friend sent me Zuckerberg's announcement of this on Instagram. Not being an Instagram user, I stupidly assumed it was a parody account and sat through the whole thing waiting for the joke.

[Anonymous Coward]

Just lovely

Super! Can't wait for delivery of my €303,933.41 edu discount DGX H100 640GB to run this valuable 8-bit quantized Llama 405B model on! I'm just so super glad I didn't waste all that money on a house or second-hand Lamborghini Aventador instead!

[hittitezombie]

Glacial

Tried with Olllama - it's glacial. It's been 5 minutes and it even have not responded to my "good morning!" message. First, mate, it's rude not to respond. Second, all you had to say was 'it's now an afternoon, hence good afternoon!"

[Kevin McMurtrie]

Re: Glacial

That size is doable. I ran LLM deepseek-coder-v2:236b on an AMD 7900x3d with 128GB RAM and a lot of swap to a fast m.2 stick. No GPU. It was tolerable at few tokens/second, but the passage of time gradually revealed that it was writing nonsense.