Meta introduces open-source AI program LLaMA 3.1 with enhanced capabilities, seen as positive for the market.

From ZDNet: 2024-07-23 13:46:00

Meta, Facebook’s parent company, has unveiled LLaMA 3.1 405B, a new open-source AI program designed to power Meta’s AI assistant across various end-user programs.

Meta CEO Mark Zuckerberg emphasized the importance of open-source AI, stating that it will benefit the world. However, questions remain about whether the new LLaMA 3.1 is truly open source.

OSI executive director Stefano Maffulli expressed concerns about the transparency and restrictions of Meta’s license, regarding the use of LLaMA 3.1 for developers.

LLaMA 3.1 features a 405 billion parameter model, boasting enhanced capabilities for natural language processing, content generation, and scientific research, positioning it competitively against leading closed models.

Meta claims that LLaMA 3.1 is not just competitive but also more cost-effective than other models like GPT-4, with estimates suggesting a 50% reduction in inference costs.

Available through various partners, LLaMA 3.1 is compatible with popular AI tools like vLLM, TensorRT, and PyTorch, enabling developers to leverage the new model for diverse applications.



Read more at ZDNet: Meta inches toward open source AI with new LLaMA 3.1