Llama 2 is Meta's open-source language model, now with 2T training tokens and up to 70B parameters. Includes pre-trained & helpfulness/safety fine-tuned versions to enable language capabilities across research & commercial use cases.
As one of the largest open-source language models out there, Llama 2 represents scale and performance. With up to 70 billion parameters trained on 2 trillion tokens - far surpassing the original Llama release - it demonstrates the power of open-source language models. 📈
🔬 As an open-source model, engineers can freely experiment to push boundaries in areas like multi-task learning, prompt programming, reinforcement learning, and beyond.
🛠️ Llama 2 provides both a 2T parameter pre-trained base model and 1M+ parameter helpfulness/safety fine-tuned versions in Llama Chat and Code Llama. It gives engineers an excellent starting point for customizing to their own use cases.
Whether through prompt programming techniques, adapter layers, decoder-only training, or full fine-tuning, the mix of model scales and specializations makes Llama 2 a versatile foundation for real-world applications. 👷♀️
🚀 State-of-the-Art Scale: Llama 2 pushes boundaries as one of the largest open-source language models, with up to 70 billion parameters trained on 2 trillion tokens.
🔬 Enabling Innovation: As an open-source model, Llama 2 fosters innovation by allowing engineers to freely experiment and build novel applications on top of its capabilities.
🌍 Accessibility: Llama 2 is designed for broad accessibility and democratizing the benefits of large language models by releasing it under a permissive license and across major cloud providers.
-
📈 Scale: With up to 70 billion parameters trained on 2 trillion tokens, Llama 2 pushes state-of-the-art scale for open-source language models. More parameters and data can enable more capable downstream applications.
-
🔬 Innovation: As an open-source model, Llama 2 enables AI engineers to build novel applications and advance research on top of its capabilities. The open ecosystem fosters innovation.
-
⚙️ Customization: Llama 2 releases include a pre-trained version and fine-tuned models for safety and helpfulness. It allows engineers to start customizing for their own use cases.
-
🌎 Accessibility: Hosted by multiple cloud providers and available via a permissive license, Llama 2 is designed for broad accessibility to empower developers and researchers globally.
-
👩💻 Community: There is great value in collaborating with the community building on top of Llama 2 - sharing ideas, best practices, benchmarks, and even additional training data.
- 👷🏽♀️ Builders: Joseph Spisak, sekyondaMeta, ruanslv, Suraj Subramanian
- 👩🏽💼 Builders on LinkedIn: https://www.linkedin.com/in/jspisak/, https://www.linkedin.com/in/surajsubramanian/
- 👩🏽🏭 Builders on X:https://twitter.com/joespeez, https://twitter.com/subramen
- 👩🏽💻 Contributors: 41
- 💫 GitHub Stars: 47.7k
- 🍴 Forks: 8.2
- 👁️ Watch: 469
- 🪪 License: LLAMA 2 COMMUNITY LICENSE AGREEMENT
- 🔗 Links: Below 👇🏽
- GitHub Repository: https://github.com/facebookresearch/llama
- Official Website: https://ai.meta.com/llama/
- Profile in The AI Engineer: https://github.com/theaiengineer/awesome-opensource-ai-engineering/blob/main/libraries/llama/README.md
🧙🏽 Follow The AI Engineer for more about llama and daily insights tailored to AI engineers. Subscribe to our newsletter. We are the AI community for hackers!
♻️ Repost this to help llama become more popular. Support AI Open-Source Libraries!