Meta Platforms has released its large language model, Llama 2, for commercial use by partnering with major cloud providers like Microsoft. Unlike charging for access, Meta is offering the model for free to other companies to benefit from improvements that arise from increased usage, testing, and issue identification.
This move not only makes Llama 2 more accessible but also positions Meta alongside other tech giants as a significant contributor to the AI arms race. CEO Mark Zuckerberg has emphasized incorporating AI advancements into all company products and algorithms as a priority.
The cost of training the models was borne by Meta, while cloud providers such as Microsoft, Amazon.com, and Hugging Face are hosting the tools and providing the computing power to run them. However, Meta has not clarified whether Microsoft will charge for access to Llama 2 via its cloud services.
The commercial rollout of Llama 2 marks the first project from Meta's generative AI group, established in February. To ensure safety, the model underwent "red-teaming exercises" conducted by Meta employees and third parties. Notably, Llama 2 is distinct from the model used for Meta's own products.
Ahmad Al Dahle, the vice president of Meta's generative AI group, praised the rapid decision-making, focused execution, and extensive investments made to bring Llama 2 to fruition, driven by over a hundred thousand requests for Llama 1.
Regarding the partnership with Microsoft, Meta sees no concerns, as Llama 2 being open-sourced enables small and medium-sized businesses to test and use large language models without the burden of hosting their custom model.
Large language models like Llama 2 power text-based chatbots and are trained with vast amounts of internet data to refine response generation processes. By open-sourcing infrastructure and AI work, Meta aims to drive industry progress and will also offer Llama 2 access to academic researchers.
Previously, Meta's history of open-sourcing includes developing PyTorch, an AI framework, which was eventually spun out to be overseen by the non-profit Linux Foundation.