Llama 2 will be available for free for both research and commercial purposes, cementing Meta’s longstanding commitment to open sourcing infrastructure and AI advancements.
The philosophy driving Meta’s decision to embrace open source stems is from its belief that innovation thrives when technology is accessible to a wide array of developers.
Additionally, the transparency afforded by open source solutions enhances safety and security, as a larger community can actively scrutinize and address potential vulnerabilities.
By open sourcing Llama 2, Meta aims to unlock boundless progress within the industry. Today, Meta unveiled pre-trained and finely-tuned Llama 2 models with parameter options of 7B, 13B, and 70B.
Remarkably, Llama 2 was trained on a dataset that surpasses Llama 1 in size by 40%, incorporating architectural advancements.
The fine-tuned models were meticulously honed with over 1 million human annotations, employing a combination of supervised fine-tuning and reinforcement learning with human feedback (RLHF).
As a result, Llama 2 sets new standards for safety and quality, achieving outstanding performance metrics.
You can directly download these models or access them via Microsoft’s Azure platform, benefiting from Microsoft’s comprehensive safety and content tools.
Full Story Source: Meta And Microsoft Collaborate To Unveil Llama 2