Meta has released an AI image generator and open source code for LLaMA 2

In the battle to win over users with artificial intelligence, another giant has broken out between Microsoft and Google. Meta wants to get a place on the podium in this frantic race and is already showing the amazing results of CM3leon (pronounced chameleon Chameleon in English). Mark Zuckerberg’s image creator also interprets them and can edit them in any way you require. Work from it or create it as per the user’s request. AI Meta Week ended with the announcement that LLaMA 2 has become open source, so anyone can download it to their computer and give it a specific app.

Its public release date is still unknown for CM3leon. Meta showed its impressive results and revealed some of its technical characteristics. “CM3leon is the first multimedia model to be trained using a recipe adapted from text-only language models,” its Meta researchers explain. The authors say that it “achieves cutting-edge performance in text-to-image generation, despite being trained at five times less computation than previous methods based on Transformers—a Google-developed system of AI language models—”.

CM3leon works from graphic representation to text and back

CM3leon also uses one of the usual methods of AI image generation, which consists of “adding a separately trained super-resolution stage to produce images with a higher resolution than the original model results”. The results displayed by the Meta are of quality, though there is no clue as to when or under what circumstances they will be publicly available.

In any case, Meta tends to open up access to its AI models a lot. On Tuesday, he and Microsoft — the largest investor in OpenAI, which brings ChatGPT’s AI to all of its commercial products — announced an agreement to make Mark Zuckerberg’s AI, LLaMa 2, available as open source available to anyone who wants it. Use it to develop applications.

The free availability of LLaMA 2 was announced during Microsoft’s Inspire event, which the Redmond company is dedicated to participating companies around the world, because its AI model will be distributed through the Azure platform, although it will also be distributed through other important platforms. such as Amazon Web Services. Meta’s move aims to make a large language model freely available to developers, unlike other closed models such as OpenAI’s GPT-4, which is only available to its partners. Meta expects the number of users to grow from this open source strategy.

LlaMA 2 is trained with 40% more data than the first Meta AI and has twice the context extension of this one.

Some of the features Meta has revealed in its remarkable AI language model are far superior to its predecessor, LLaMA 1. The new feature is trained with 40% more data than the first and has twice the context extension of that model.

LLaMA 2 is now available to anyone with three different sizes that are measured according to the number of parameters used in the language model: 7,000, 13,000, and 70,000 million. This latest version is comparable to GPT-3.5, the model with which OpenAI launched ChatGPT last November, although it now also has a much improved model, of which few characteristics are known, GPT-4, which has the potential to go deeper. More in his answers and multimedia, able to use text and images.

Meta guarantees that its AI models have “passed both internal and external security testing.” The company also provides developers with a guide on responsible practices and security assessments. The European Union is preparing a law that regulates this technology.

Read also