Meta Announces LLaMA 2, with plans for free access for both research and commercial usage
July 18, 2023 - Palo Alto, California
Meta announced the official release of their open source large language model, LLaMA 2, for both research and commercial use, marking a potential milestone in the field of generative AI.
Meta AI Research (FAIR) is helmed by veteran scientist, Yann LeCun, who has advocated for an open source approach to AI and LLM development. This puts them at odds with the most recent release from OpenAI, which controversally hid major details of the architecture in the whitepaper for their GPT-4 model, for the first time in their history of GPT-related releases.
This release includes model weights and starting code for pretrained and fine-tuned LLaMA language models, ranging from 7 billion to 70 billion parameters. The pretrained models have been trained on an extensive dataset of 2 trillion tokens, offering double the context length compared to LLaMA 1. Additionally, the fine-tuned models have been trained on over 1 million human annotations, further enhancing their performance and accuracy.
To learn more about LLaMA 2 and its capabilities, as well as register to download the model, visit the official LLaMA website. Access is gated via a submit form, and requires acceptance of their terms.