Meta release Llama 3

Meta has officially released Llama 3, a new set of open artificial intelligence (AI) models that includes several enhancements and features designed to improve AI applications on different platforms. Llama 3 models are available in two sizes: 8 billion parameters and 70 billion parameters, with both pre-trained and instruction-trained versions available.

Key features and integrations

• Model variants: Llama 3 is available in two main variants-a basic version and an instruction-fitted version for each of the two sizes (8B and 70B). This allows greater flexibility in deployment according to specific application needs.
• Integration and accessibility: Llama 3 models are integrated into various platforms, including Google Cloud Vertex AI and Azure AI, making them accessible to a wide range of developers and applications.
• Advanced tokenizer and training framework: The models feature an improved tokenizer and are trained on custom server clusters, improving efficiency and performance.

Applications and Uses

• Meta’s AI Assistant: Llama 3 powers the new version of Meta’s AI assistant, an intelligent assistant available on Meta’s various platforms such as Facebook, Instagram, WhatsApp and Messenger. This assistant is designed to handle a variety of tasks, from processing images to providing real-time information.
• Enhanced chatbot features: The new chatbot features introduced with Llama 3 aim to provide more accurate and diverse responses, improving user interaction on social media.

Open source availability and commitment

• Open source access: In line with its commitment to open AI, Llama 3 models are freely available under permissive licenses, allowing developers to use and modify them for both research and commercial purposes.
• Security and ethical considerations: Meta has taken steps to ensure responsible use of Llama 3, providing guidelines and tools for developers to implement the models safely in their applications.

Future developments

• Expansion plans: Meta has planned to introduce larger models with more than 400 billion parameters in the future, with the goal of further improving the model’s capabilities and expanding its applicability to more complex tasks.

The release of Llama 3 is an important step in the continuation of Meta’s efforts to develop AI technology and make powerful tools available to the technology community.

How to try Llama 3 and other models on a notebook

Llama 3 is available in two dimensions: 8 billion parameters and 70 billion parameters. Using LM Studio allows you to automate the complex steps required to test a model locally, without the need to go through the Meta AI site. This free tool is compatible with macOS, Windows and Linux, and models can be downloaded from the various data

LM Studio Features.

LM Studio allows you to run very large models by loading them entirely into memory, as well as taking advantage of Metal’s acceleration that allows you to use the 8-billion-parameter Llama 3 even on a Mac M1 with 8 GB of RAM. The interface is super easy to use: once installed, a series of suggested models will appear, but you can use a convenient search to browse through all the models on Hugging Face’s Model Hub.

Benefits of LM Studio

With LM Studio it is possible to try almost all LLM text models available in open format, including Meta’s recent LLama 3 models. The 8-billion-parameter version requires only 3.5 GB of disk space and a similar amount of memory, while the 70-billion-parameter version requires about 35 GB of space and an amount of memory not usually found on consumer notebooks.