The Llama 3.1 collection of multilingual LLMs, which includes sizes of 8B, 70B, and a novel 405B parameter, was recently released by Meta. With this, the Llama 3.1-405B, which is available on IBM® Watsonx.AITM, becomes the biggest open-source language model currently available. These models can be set up in on-premises, hybrid, or IBM cloud environments.
This release follows the April launch of Llama 3 models and reflects Meta’s ongoing efforts to enhance multilingual and multimodal capabilities, extend context lengths, and improve overall performance, including reasoning and coding.
Advancing Open and Responsible AI Innovation
In December 2023, Meta and IBM formed the AI Alliance with over 50 global partners to shape AI development responsibly. With over 100 members, the AI Alliance promotes safe, secure, diverse, and competitive AI innovation. It supports projects focused on benchmarks, addressing societal challenges, building AI skills, and encouraging safe and open AI development.
By offering cutting-edge open-source models and development tools to assist the global AI community in responsibly innovating, Llama 3.1 supports this mission. This release includes guardrails to standardise trust and safety tools for generative AI, cybersecurity evaluations, and safety measures.
Exclusive: Meta just released Llama 3.1 405B — the first-ever open-sourced frontier AI model, beating top closed models like GPT-4o across several benchmarks.
I sat down with Mark Zuckerberg, diving into why this marks a major moment in AI history.
Timestamps:
00:00 Intro… pic.twitter.com/wI0X86P0dM— Rowan Cheung (@rowancheung) July 23, 2024
Llama 3.1-405B vs. Leading Models
The Llama 3.1 models, especially the new 405B version, match or exceed the performance of leading closed-source models. For example, in knowledge tests, the Llama 3.1-405B scores 87.3%, outperforming models like OpenAI’s GPT-4-Turbo and Google’s Gemini 1.5 Pro.
Benefits Beyond Performance
The 405B model offers more than just high performance. As an open-source model, it provides stability and control for researchers and enterprises, allowing modifications and consistent reproducibility, unlike many closed-source models.
Using Llama 3.1-405B
IBM and Meta believe open models like Llama 3.1 foster better products, quicker innovation, and a healthier AI market. The 405B model’s size offers unique opportunities for synthetic data generation, knowledge distillation, and domain-specific fine-tuning. It can also be used as a benchmark to judge the quality of other models.
To make the most of Llama 3.1-405B, Meta recommends using a platform like IBM® watsonx.aiTM, which offers robust features for model evaluation, safety, and retrieval-augmented generation (RAG).
Upgrades Across All Sizes
While the 405B model is the highlight, all Llama 3.1 models come with significant upgrades, including:
-
Longer Context Windows: Llama 3.1 models have a context length of up to 128,000 tokens, a significant increase from the previous 8,192 tokens. This allows for longer conversations and the ability to handle larger documents and code samples.
-
Multilingual Support: Llama 3.1 models now support multiple languages, including Spanish, Portuguese, Italian, German, and Thai, with more languages in validation.
-
Optimised Tool Use: The instruction-tuned models are optimised to interface with various tools for search, image generation, code execution, and mathematical reasoning.
Getting Started with Llama 3.1
Meta’s Llama 3.1 release offers a great opportunity to customise state-of-the-art AI models for your specific needs. IBM watsonx provides the tools to deploy, fine-tune, and integrate these models into your business environment. The 405B model is available today, with the 8B and 70B models coming soon.
Try Llama 3.1-405B in watsonx.ai™
Start with RAG tutorials using Llama 3.1-405B and watsonx.ai:
-
No-code RAG
-
RAG with PDFs
-
RAG with web data
Inputs from Multiple Agencies
Media from multiple sources(X)
Ⓒ Copyright 2024. All Rights Reserved Powered by Vygr Media.
