Gaming

Meta launches Llama 3.1, an open source AI model that seeks to compete with GPT-4o

Image

Meta’s commitment to large-scale language models is distinguished by being open source, which makes it easy to earn the title of biggest and best in its sector. However, the company led by Mark Zuckerberg does not want to stop there, so it has Llama 3.1 presentedthe next generation of its AI model that “rivals the best closed-source models.” Meta refers to OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet, which are backed by Microsoft and Amazon, respectively.

“Llama 3.1 405B is the first open source model to compete with leading AI models when it comes to the most advanced features of general knowledge, driving ability, mathematics, tool usage, and multilingual translation,” says Meta. This new version of the model also allows the Llama-based Meta AI assistant to be available in new countries and languages, including Mexico and several South American territories such as Argentina, Peru, and Chile.

Meta AI’s expansion will allow users to use this model (where available) in Spanish via Facebook, Instagram, Messenger, WhatsApp and on the meta.ai website. Meta announced a few days ago that it does not intend to launch its most advanced AI models in Europe due to the “unpredictable regulatory environment.”

Meta’s benchmarks say Llama 3.1 outperforms GPT-4, GPT-4o, and Claude 3.5 Sonnet on some tasks.

Llama 3.1 is a significantly more complex model than the Llama 3 models seen so far. Its largest version has 405 billion parameters or modifiable elements and has been trained with more than 16,000 Nvidia H100 GPUs, which according to Meta makes this model the first trained at this scale. The smaller Llama 3.1 models total 8 and 70 billion parameters. All three are multilingual, have a context length of 128K, and their ecosystem is configured and ready to work on services from Amazon Web Services, Microsoft Azure, Google Cloud, and other partners.

Meta is confident that Llama’s open source nature will appeal to developers who want to integrate it into their products. Llama is a model that can be downloaded, customized to suit needs and applications, trained with new data sets, and further fine-tuned. “Developers can fully customize their applications and run them in any environment, whether on-premises, in the cloud, or even locally on a laptop, all without sharing data with Meta,” the company says. In addition, Meta argues that Llama offers one of the lowest cost per application in the world. token lowest in the sector.

Source link