What is Llama 2: the new Meta and Microsoft AI model

Llama is the Artificial Intelligence language developed by Meta together with Microsoft, a rival to other generative AIs such as GPT-4 from ChatGPT or LaMDA behind Google Bard. Llama 2 is an open source model and this second generation is much more advanced than the first.

What is Llama 2?

Some basic facts:

• This LLM (Large Language Model) has been trained on 70 billion parameters, 2 trillion tokens, and has twice the context length of Llama 1.
• Its fitted models have been trained on more than 1 million comments and observations made by humans.
• It works in 27 languages, although the quality differs quite a lot from one language to another, with English winning.
• Llama is a pre-trained model, which means that it works with data updated until March 2023 and does not connect to the internet to search for new information or context. That is, if, for example, you ask it about the colossal success of the Barbie movie... it won't know what you are talking about.
• It would have a performance equivalent to something like the GPT-3.5 of ChatGPT in its no-cost version.
• Llama 2 is powered by Azure, Microsoft's public cloud platform, and will also be available through Amazon Web Services (AWS), Hugging Face, and other providers. In other words, companies, organizations, and developers will be able to integrate this language model into their products and projects using Microsoft or Amazon services.
• It is a free service that anyone can try, although for the moment Meta has not created a specific application for it. We will explain how you can use it.

Who can use Llama 2?

As mentioned, Llama 2, in its 7B, 13B, and 70B parameter versions, can be used by any user, for research and commercial purposes. Meta insisted on the open approach of its AI, justifying it as follows:

"Opening access to today's AI models means a generation of developers and researchers can stress test them, identifying and solving problems fast, as a community. By seeing how these tools are used by others, our own teams can learn from them, improve those tools, and fix vulnerabilities."

But even if you are not a researcher or a company, it is still possible to use this LLM. See below to find out how.

How to download and use this AI?

Using Llama 2 requires completing an application process to download it to Windows so that you can use this tool on your computer. To access it, complete the following steps:

1. Go to the official website and click on the Download the Model button.
2. Next, fill in your personal details: first and last name, email, country of residence, and the name of your organization.
3. Scroll down to the bottom of the form, click on Accept the terms and conditions, and then click on Accept and Continue.
4. Meta will process your request and, if accepted, will send you the links and files you need to download to your computer and start using it.

Note: the waiting time can be several days or weeks, so... be patient.

How can you try Llama 2 chat?

In case you are a user with no investigative or commercial interest in the tool but don't want to wait, the good news is that there are several ways to try out scaled-down versions of this AI. These are chatbots that don't have the more advanced features of other chatbots, such as history or prompts, but they will give you an idea of how it works:

• Developer Pietro Schirano has created an online app with Replit and Replicate, which gives you the option to try the Llama 2 chat directly, for free and without having to register.
• HuggingFace, one of the partner providers, also gives you the possibility to test Llama 2 from the same blog post in which they announce their partnership with Meta and Microsoft to implement this LLM. In particular, you will be able to try a demo of the 70b-chat-hf model, which, as you will see, is somewhat different from Schirano's model.

Go ahead and try it out and let us know how it goes!