Jul 23, 2024 at 8:15 PM Meta unveils Llama 3.1 open-source AI models, including the 405B with 128K context length
Meta has introduced the Llama 3.1 family of AI models, which includes Llama 3.1 8B, Llama 3.1 70B, and the massive Llama 3.1 405B. These models feature a 128K context length, enhancing their processing capabilities. Meta's evaluation of these models involved over 150 benchmark datasets and several human evaluations, comparing their performance with other leading models in real-world scenarios.
The Llama 3.1 405B model is positioned as competitive with top foundation models like GPT-4 and Claude 3.5 Sonnet, and is described as the largest and most capable openly available foundation model. The smaller models, Llama 3.1 8B and Llama 3.1 70B, also perform well against both closed and open models of similar sizes.
Despite high development costs, Meta's AI models remain open-source, requiring license approval only from companies with hundreds of millions of users, highlighting the potential for open-source AI to surpass proprietary models from Microsoft, OpenAI, Anthropic, and Google. The model weights are available for companies to train and customize according to their needs, with access provided through platforms like AWS, NVIDIA, Databricks, Groq, Dell, Azure, and Google Cloud, with specific availability of Llama 3.1 405B via Azure AI's Models-as-a-Service as a serverless API endpoint.
Can't believe the Zuck is doing more to open source AI than OpenAI