Ref: https://indianexpress.com/article/explained/explained-sci-tech/what-is-llama-3-metas-most-sophisticated-and-capable-large-language-model-9280994/

Introduction:

Llama, also known as Large Language Model Meta AI, represents a series of LLMs (Large Language Models).

Unveiled by Meta AI in February 2023.

• The Large Language Model (LLM) Meta AI refers to a family of advanced language models developed by Meta AI. 
• These models are designed to understand and generate human-like text based on vast amounts of data they have been trained on. 
• LLMs are capable of various language-related tasks such as text generation, translation, summarization, and question answering.

Initially, Llama debuted in four variations: 7B, 13B, 33B, and 65 billion parameters. 

Parameters indicate the size and complexity of an AI model, with higher numbers suggesting greater sophistication and capability. 

What is Llama 3?

• Meta has introduced Llama 3, the newest version of its LLM, boasting enhanced performance and AI capabilities.
• The instruction-tuned version is specifically aimed at empowering AI chatbots for engaging in conversations with users.
• It specialises in areas such as reasoning, code generation, and following instructions, rendering it more adaptable and versatile.
• It is speculated that, it can generate diverse forms of text such as poems, code, scripts, and musical compositions. 
• Additionally, it may be capable of summarising factual topics and translating languages.

How to try?

• Meta has integrated Llama 3 into its Meta AI platform. This allows users to access it across various Meta-owned platforms such as Facebook, Instagram, WhatsApp, Messenger, and the web.
• Developers can readily access Llama 3 through Meta’s integration with the Hugging Face ecosystem, a popular platform for AI model development and deployment.
• It is also accessible via other platforms including Perplexity Labs, Fireworks AI, and major cloud provider platforms such as Azure ML and Vertex AI.
• Meta plans to make Llama 3 models available on additional platforms in the near future, including AWS, Google Cloud, Hugging Face, Databricks, Kaggle and more.
• Currently, Meta AI with Llama 3 is available in English across the US on WhatsApp.
• It plans to expand to countries like Australia, Canada, Ghana, Jamaica, Malawi, New Zealand, Nigeria, Pakistan, Singapore, South Africa, Uganda, Zimbabwe, and Zambia.

Comparison between Llama 2 and Llama 3

Feature	Llama 2	Llama 3
Release Date	July 18, 2023	April 18, 2024
Model Sizes	7B, 13B, 70B parameters	8B, 70B parameters
Training Data	40% more data than Llama 1	7x times more data than Llama 2
Context Length	Double the context length of Llama 1	Not specified, but likely improved
Fine-tuning	Tuned on human preference data	Focus on careful curation of instruction-tuning data
Performance Benchmarks	Outperformed other open-source models	Claimed by Meta to be “best open-source models of their class”
Open Source	Yes	Yes (with limitations for high user base platforms)
Availability	Research and Commercial Use	Research and some Commercial Use (<700 million monthly active users)
Focus	Natural Language Understanding	Improved alignment an