LLAMA 3 (Syllabus: GS Paper 3 – Sci and Tech)

News-CRUX-10     23rd April 2024        
Samadhaan

Context: Meta recently introduced its most capable Large Language Model (LLM), the Meta Llama 3.


Llama 3

  • About: It is a family of Large Language Model (LLM) introduced by Meta AI.
  • LLMs: A large language model is a type of artificial intelligence algorithm that uses deep learning techniques and massively large data sets to understand, summarize, generate and predict new content.

oThe term generative AI also is closely connected with LLMs, which are, in fact, a type of generative AI that has been specifically architected to help generate text-based content.

  • Model Sizes: Similar to its predecessors, Llama 3 comes in various sizes, with the initial release offering models with 7B, 13B, 33B, and 65 billion parameters.
  • Superior Performance: The 13B version of Llama 3 reportedly surpassed the performance of OpenAI's GPT-3, despite having fewer parameters, showcasing its efficiency and effectiveness in natural language processing tasks.
  • Release of Llama 2: Before Llama 3, Meta AI introduced Llama 2 in July of the preceding year, representing a significant advancement over its predecessor.
  • Features of Llama 2: It was available in 7B, 13B, and 70B parameter configurations, indicating a broader range of options for users. Moreover, it was trained on a dataset expanded by 40% compared to the original Llama model, resulting in improved performance and capabilities.

How good is Llama 3?

  • Meta's Assertion: Meta claims that Llama 3, with its 8B and 70B parameter models, represents a significant advancement over Llama 2.
  • Expanded Capabilities: Llama 3 exhibits notable enhancements in various capabilities, including reasoning, code generation, and instruction following.
  • Increased Steerability: The incorporation of advanced post-training processes has made Llama 3 more adaptable and easier to control, offering users greater flexibility and control over its functionalities.
Samadhaan