Jul 24, 2024 10:49:00

Meta releases Llama 3.1, an AI model that surpasses GPT-4

Meta released the large-scale language model ' Llama 3.1 ' on July 23, 2024. Llama 3.1 is open source and is said to have performance equal to or better than cutting-edge closed-source AI models such as GPT-4 and GPT-4o.

Llama 3.1

https://llama.meta.com/

Introducing Llama 3.1: Our most capable models to date
https://ai.meta.com/blog/meta-llama-3-1/

Llama 3.1 offers models with parameter counts of 405 billion, 70 billion, and 8 billion, all with 128,000 context windows.

Below is a table comparing the benchmark results of 'Llama 3.1 405B', which has 405 billion parameters, with 'Nemotron 4 340B Instruct', 'GPT-4', 'GPT-4o', and 'Claude 3.5 Sonnet'. Llama 3.1 405B recorded scores higher than GPT-4 and GPT-4o in multiple tests.

Furthermore, the results of a human comparison of the output results of 'Llama 3.1 405B' with 'GPT-4', 'GPT-4o', and 'Claude 3.5 Sonnet' are shown below. It can be seen that more than half of the subjects rated the output results of Llama 3.1 405B as equivalent to or better than the three models.

The benchmark results for the 'Llama 3.1 8B' and 'Llama 3.1 70B' look like this. Both models recorded higher scores than models of the same size.

A demo of a chat AI using 'llama-3.1-405b-instruct' was already available on NVIDIA's AI platform '

NVIDIA NIM ,' so I tried it out. First, I asked in Japanese, 'Which should I choose: wired earphones or completely wireless earphones?' and it explained how to choose by listing the advantages and disadvantages of both.

Next, I asked, 'How do I display line numbers in Vim?' and he taught me the command to display line numbers and how to write it in the configuration file.

Llama 3.1 is already available for free download on Meta's official website and Hugging Face. It's also released under the LLAMA 3.1 COMMUNITY LICENSE AGREEMENT , which allows for redistribution and commercial use under certain conditions.

Llama 3.1
https://llama.meta.com/

Llama 3.1 - a meta-llama Collection
https://huggingface.co/collections/meta-llama/llama-31-669fc079a0c406a149a5738f

In addition, Meta CEO Mark Zuckerberg announced Llama 3.1 in a statement, saying, 'Open source is essential for a bright future for AI.' 'The Llama project is open sourcing to improve efficiency, optimize processing chips, and develop the ecosystem, and we believe that open sourcing will not diminish its technical merit.' 'Until now, Llama has been an internal project for Meta. With Llama 3.1, we are forming an internal team to make it available to as many developers and partners as possible.'