In the world of artificial intelligence and machine learning, language models have taken center stage, shaping the way we think about technology and the internet. Among these models, LLaMA has emerged as an interesting player, promising both innovation and high accessibility.
Large Language Model Meta AI, or LLaMA for short, is a family of large language models developed by Meta AI. It made its debut in February 2023, boasting four different model sizes, ranging from 7 billion to 65 billion parameters. What set LLaMA apart was its exceptional performance on natural language processing benchmarks, often outshining models with significantly more parameters.
Although Meta introduced Llama 2 as the next generation of their Open Source large language model, the LLaMA 2 license is far from being Free or Open Source. The LLaMA license, while much more permissive than licenses attached to many commercial machine learning models, comes with certain restrictions that prevent it from being classified as a true Free or Open Source license. Notably, companies with over 700 million monthly active users must obtain a commercial license from Meta AI. Moreover, users are prohibited from using LLaMA outputs to train non-LLaMA-based models, and all usage must adhere to Meta’s use policy, which restricts certain types of outputs.
One might think this a small semantic issue concerning only tech giants like Google or Amazon since in many aspects the LLaMA’s license may appear quite similar to some Permissive Free and Open Source licenses. However, the key difference lies in the underlying ideology.
Nothing promotes user adoption faster than free software, free to use and preferably free of money. Most likely this is why the LLaMA license resembles Free and Open Source licenses. In contrast, the LLaMA’s license limitations are more aligned with a commercial model and its underlying ideology and it’s crucial to recognize that it may evolve over time.
Changing the license of a project open for third-party contribution is not always simple, in some cases the approval of all the contributors is required, making it virtually impossible in large projects like the LINUX kernel. Although the LLaMA project is open to contributions by Individuals and Companies, contributions are subject to signing a Contributor License Agreement (CLA) granting Meta a wide license in the contribution, so changing the LLaMA project’s license will not cause any legal issues and can be done at any given time.
Understanding the nuances between “open source” and “open source-inspired” licenses is essential for businesses and developers looking to leverage the LLaMA project. While the LLaMA’s license currently appears more permissive than most free commercial software, we should be prepared for potential changes in the future, especially if LLaMA continues to gain ground in the landscape of AI language models.