Llama 3Translation site

2wks agorelease 880,565 0 95.1K
Collect 0

Meta introduces Llama 3, offering models with 8B and 70B parameters, marking a significant advancement in open-source AI. Llama 3 builds upon its predecessors' strengths...

Location:
United States
Language:
US
Collection time:
2025-05-30
Open siteMobile view
Llama 3Llama 3
Llama 3Open site

Meta has recently released the next-generation open-source large language model, Llama 3, offering models with 8B (8 billion parameters) and 70B (70 billion parameters), marking another significant advancement in the field of open-source artificial intelligence.

  • Multiple Parameter Scales: Offers models with 8B and 70B parameters to meet various application needs.
  • Efficient Training: Pre-trained on over 15 trillion tokens, including a large amount of high-quality data.
  • Improved Architecture: Utilizes a more efficient tokenizer and Grouped Query Attention (GQA) technology to enhance inference efficiency and long-text processing capabilities.
  • Multilingual Support: Pre-training data includes high-quality non-English data from over 30 languages, enhancing multilingual capabilities.
  • Performance Enhancement: Demonstrates outstanding performance in multiple benchmark tests, surpassing other models of the same scale.
  • Enhanced Security: Introduces new trust and security tools to improve the model’s safety and reliability.
  • Openness: Provides model weights and code, supporting researchers and developers in customization and fine-tuning.

    • Relevant Navigation

    标签云