Meta Platforms, in collaboration with Microsoft, has released an updated version of its generative language model, LLaMA (Large Language Model Meta Ai). LLaMA 2 aims to provide start-ups and companies with the ability to develop customized software using the open-source language model. Meta claims that LLaMA 2 was trained using 40 percent more data than its predecessor, LLaMA 1, and includes twice as much context. The latest version of the model is available in three different sizes, with varying numbers of parameters.
According to Meta, LLaMA 2 has outperformed other open-source language models of a similar size on various benchmarks. The company plans to launch a commercial variant of LLaMA in the near future, as reported by the Financial Times. However, LLaMA 2 is currently available for free, even for commercial applications. This free accessibility is expected to benefit users while also enhancing security and promoting further development.
Meta emphasizes the importance of developers and researchers testing and identifying any issues with LLaMA 2 so that they can be addressed promptly. By observing how these tools are utilized by others, Meta’s teams can learn and improve upon them. The demand for LLaMA 1 has exceeded expectations, with over 100,000 requests, leading to the development of LLaMA 2.
LLaMA 2 is now accessible through Microsoft’s Azure AI Model Catalogue, as well as other platforms such as AWS (Amazon Web Services) and Hugging Face. Before using the language model, users are advised to read the Acceptable Use Policy, the Responsible Use Guide, and an in-house research paper that provides insights into known issues and planned workarounds.
In an effort to encourage collaboration and knowledge sharing, Meta is launching a partnership offer for AI researchers at universities worldwide. They also plan to initiate the Llama Impact Challenge, a competition where participants can utilize LLaMA 2 to address significant challenges related to areas such as the environment and education.
