Meta launches Llama 3 !
Meta has just launched Llama 3.
Summary:
Here's 10 key points about Meta's Llama 3 large language model (LLM):
New Version, More Powerful: Llama 3 is the latest iteration of Meta's LLMs, boasting significant advancements compared to previous versions and other open models like Mistral and Gemma.
Foundation for Meta's AI Assistant: The model's reliability and accuracy are showcased in its use as the underlying technology for Meta's new AI assistant, available across various Meta platforms.
Multiple Model Sizes and Variants: Llama 3 offers four models (base and instruction-tuned) in two sizes (8 billion and 70 billion parameters), catering to different needs.
8 Billion Parameters: Pre-trained LLM for routine tasks.
70 Billion Parameters: Instruction-tuned for coding and multi-language applications. The instruction-tuned versions are designed for tasks like chatbots.
Improved Capabilities: Llama 3 supports a longer context length (8,000 tokens) compared to previous models, allowing for more complex interactions. It also integrates with the Hugging Face ecosystem for easier adoption by developers.
Safety Focus: Meta released Llama Guard 2 alongside Llama 3, specifically designed to improve safety and reliability in production use cases.
Impressive Benchmarks: The 70 billion parameter model outperforms other prominent models like GPT-3.5 and Gemini in various tasks like coding, writing, and summarization.
Extensive Training: Trained on a massive 15 trillion token dataset, seven times larger than Llama 2, contributing to its improved performance.
Future Advancements: Meta is actively developing even larger models (over 400 billion parameters) with multi-language and multi-modality support.
Open Access with Commitment: Meta emphasizes its commitment to open access by offering Llama 3 for free, fostering innovation and widespread testing. It's important to note that Meta differentiates Llama 3 as "openly accessible" rather than "open source" since they haven't released the source code or training data.
Hardware Optimization: The models are optimized for hardware from major players like Intel, AMD, and Nvidia.
Overall, Llama 3 represents a significant advancement in open LLMs, with potential for broad adoption due to Meta's reach and industry partnerships.
Try Meta AI
Meta’s AI assistant is now built with Llama 3 technology.
You can now use Meta AI on Facebook, Instagram, WhatsApp, Messenger, and the meta.ai website.
More information at https://ai.meta.com/blog/meta-llama-3/