Meta Llama 3 is Released

Meta unveils its most powerful open-source large model, Llama 3, reclaiming its position as the leader in open-source large models overnight.

On April 18, local time, Meta released two open source Llama 3 8B and Llama 3 70B models for free use by external developers. Both versions of Llama 3 will also be available on major cloud providers soon.

GitHub Project address: https://github.com/meta-llama/llama3

Llama 3 is out

According to Meta, Llama 3 8B and Llama 3 70B are currently the best-performing open source models of the same size. Llama 3 8B performs better than llama 2 70B on some test sets! And in the next few months, Meta will launch more versions. NVIDIA senior scientist Jim Fan believes that the versions of Llama 3-400B and above that may be released in the future will become a kind of "watershed", and the open source community may be able to use GPT-4 level models.

After the release of Llama 3, Meta founder Zuckerberg told the media, "Our goal is not to compete with open source models, but to surpass everyone and build the most leading artificial intelligence." Specifically, the highlights of Llama 3 and features summarized below:

Based on training of more than 15T tokens, the size is more than 7 times that of the Llama 2 data set;

Training efficiency is 3 times higher than Llama 2;

Supports 8K long text, and the improved tokenizer has a vocabulary of 128K tokens for better performance;

State-of-the-art performance on a wide range of important benchmarks;

Enhanced reasoning and coding abilities;

There is a major breakthrough in security, and the new trust and security tools with Llama Guard 2, Code Shield and CyberSec Eval 2 can also have better "error rejection performance" than Llama 2.

Now, interested users can experience Llama 3 on HuggingChat.

HuggingChat with Llama 3 model

Address: https://huggingface.co/chat/ (need to manually switch to model Llama 3)