Meta unveils new Llama 3 models after nearly nine months, now open-source under Meta’s Community License Agreement. The latest lineup includes Llama 3 in two sizes: 8B parameters and 70B parameters, featuring an 8K token context length.
Meta is currently training its largest multimodal model, boasting 400 billion parameters, set to debut soon. However, it remains uncertain whether Meta will opt to open-source this sizable model upon its public release.
Currently, Meta is incorporating its Llama 3 models into its entire suite of social media platforms, spanning Facebook, Instagram, WhatsApp, Messenger, and the web. Similar to ChatGPT, users can engage with Llama 3 models through meta.ai (visit), free of charge. Additionally, Meta AI enhances user experiences across feeds, chats, and search functions within the apps.
Utilize Meta AI to instantly generate images and GIFs, access up-to-date information on any topic (leveraging both Google and Bing), ask questions directly from your feed, and much more. Explore our firsthand account of using Meta AI on WhatsApp.
Meta is introducing AI features across its social media applications and debuting the web portal in English across multiple regions, including Australia, Canada, Ghana, Jamaica, Malawi, New Zealand, Nigeria, Pakistan, Singapore, South Africa, Uganda, Zambia, and Zimbabwe. The company plans to expand to additional countries in the near future.
Regarding the Llama 3 models, Meta asserts that its 8B model surpasses Google’s Gemma 7B and Mistral 7B models. Additionally, the Llama 3 70B model narrowly outperforms Gemini 1.5 Pro and exhibits superior performance compared to Claude 3 Sonnet in the MMLU benchmark.
Extensive testing is required to determine the standing of Meta’s models against competitors. However, initial impressions suggest that Meta has introduced robust models with the Llama 3 lineup. What are your thoughts on Meta’s new models? Share your opinions in the comments below.
0 Comments