Meta Llama 3

#1

Post by bratao » Thu, Apr 18, 2024, 3:57 PM UTC

Meta Llama 3
llama.meta.com

Re: Meta Llama 3

#2

Post by gigel82 » Thu, Apr 18, 2024, 4:05 PM UTC

Can't wait to play with this in llama.cpp; the 8B results look impressive.

Re: Meta Llama 3

#3

Post by jelled » Thu, Apr 18, 2024, 4:06 PM UTC

Looks like 8K context length. Seems to compare well against Gemini Pro 1.5 and Claude 3 Sonnet according to the included benchmarks.

Re: Meta Llama 3

#4

Post by bbig » Thu, Apr 18, 2024, 4:06 PM UTC

They've got a console for it as well, https://www.meta.ai/

And announcing a lot of integration across the Meta product suite, https://about.fb.com/news/2024/04/meta-ai-assistant-built-wi...

Neglected to include comparisons against GPT-4-Turbo or Claude Opus, so I guess it's far from being a frontier model. We'll see how it fares in the LLM Arena.

Re: Meta Llama 3

#5

Post by minimaxir » Thu, Apr 18, 2024, 4:07 PM UTC

The model card has the benchmark results relative to other Llama models including Llama 2: https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md...

The dramatic performance increase of Llama 3 relative to Llama 2 (even Llama 2 13B!) is very impressive. Doubling the context window to 8k will open a lot of new oppertunities too.

Re: Meta Llama 3

#6

Post by DenseComet » Thu, Apr 18, 2024, 4:07 PM UTC

Interesting to see that their model comparisons don’t include OpenAI models.

Re: Meta Llama 3

#7

Post by namanyayg » Thu, Apr 18, 2024, 4:07 PM UTC

I'm so surprised that Meta is actually leading the open source AI landscape?! I've used llama2 extensively and can't wait to try out llama3 now. I can't believe that it does better than Claude 3 in benchmarks (though admittedly claude 3 seems to have been nerfed recently)

I sure do wish there was more info about how its trained and its training data.

Re: Meta Llama 3

#8

Post by drag0s » Thu, Apr 18, 2024, 4:08 PM UTC

> We’re rolling out Meta AI in English in more than a dozen countries outside of the US. Now, people will have access to Meta AI in Australia, Canada, Ghana, Jamaica, Malawi, New Zealand, Nigeria, Pakistan, Singapore, South Africa, Uganda, Zambia and Zimbabwe — and we’re just getting started.

Re: Meta Llama 3

#9

Post by tosh » Thu, Apr 18, 2024, 4:09 PM UTC

* 8k context length

* Llama3 8B beats Mistral 7B

* Llama3 70B beats Claude 3 Sonnet and Gemini Pro 1.5

Re: Meta Llama 3

#10

Post by CuriouslyC » Thu, Apr 18, 2024, 4:09 PM UTC

DenseComet wrote: post #6

Interesting to see that their model comparisons don’t include OpenAI models.

Maybe not the reason, but claude sonnet obliterates gpt3.5 and there isn't a direct llama competitor to gpt4.