Introduction: On April 18, Meta released two open source Llama 3 8B and Llama 3 70B models for free use by external developers. This release is not only a showcase of technological innovation, but also a manifestation of the spirit of open source, indicating that a new chapter in AI applications is about to begin.
This time, Meta has open sourced two models, Llama 3 8B and Llama 3 70B, with two versions: pre-training and instruction fine-tuning respectively.
Platform link
Visit the link below and enter the platform to experience it (invitation code: llama3):
https://ai.gitee.com/hf-models/meta-llama/Meta-Llama-3-8B-Instruct
Maximum 400 billion parameters
Meta Llama 3 introduces a number of technical innovations, including support for longer context windows, multiple model sizes, and improved performance. In addition, the new model has performed at the top level in multiple industry benchmarks, especially its improvements in logical reasoning and programming instruction following, making it unique among similar open source models.
The highlights and features of Llama 3 are as follows:
-
Based on training with more than 15T tokens, the size is more than 7 times that of the Llama 2 data set;
-
The training efficiency is 3 times higher than Llama 2;
-
Supports 8K long text, and the improved tokenizer has a vocabulary of 128K tokens to achieve better performance;
-
State-of-the-art performance on a wide range of important benchmarks;
-
Enhance reasoning and coding skills;
-
There is a major breakthrough in security. The new trust and security tools with Llama Guard 2, Code Shield and CyberSec Eval 2 can also have better "error rejection performance" than Llama 2.
Sustainability and community engagement
Meta Llama 3 is more than just a technology product, it also represents Meta's commitment to continued development. By making this model open source, Meta hopes to stimulate the innovative potential of the global developer community and jointly promote the advancement of artificial intelligence technology. This move not only promotes the popularization of the technology, but also ensures that community feedback is fully considered during the development process.
Forward thinking on safety and ethics
In developing Meta Llama 3, Meta paid special attention to safety and ethical responsibility. Newly introduced tools such as Llama Guard 2 and CyberSec Eval 2 provide developers with more resources to use and deploy these models responsibly. This demonstrates Meta's industry leadership in ensuring safe and reliable deployment of AI.
The release of Meta Llama 3 is not only a technological advancement, but also a manifestation of the spirit of open innovation. It not only provides developers with powerful new tools, but also sets new standards for the future development of the global AI ecosystem. In the wave of AI, Meta has demonstrated the power of cooperation and innovation through Llama 3, and hopes that this platform will inspire more innovative applications of artificial intelligence.