The Llama 3 from Meta is the latest version of the open-access Llama series, now available on the Hugging Face platform. Seeing Meta's continued commitment to the development of the open AI field is exciting, and we are very happy to fully support this release and achieve deep integration with the Hugging Face ecosystem.
Llama 3 offers two versions: the 8B version is suitable for efficient deployment and development on consumer-grade GPUs; the 70B version is designed for large-scale AI applications. Each version includes basic and instruction-tuned forms. Additionally, a new version of Llama Guard based on Llama 3 8B fine-tuning has also been released as Llama Guard 2 (secure fine-tuning version).
We work closely with Meta to ensure that their products can seamlessly integrate into the Hugging Face ecosystem. On the Hub, you can find these five open-access models (including two basic models, two fine-tuned models, and Llama Guard).
The key features and functionalities of this release include:
- Models on the Hub, with model cards and licensing information
- 🤗 Transformers integration
- Hugging Chat integration for Meta Llama 3 70B
- Inference functionality integrated into inference endpoints, Google Cloud, and Amazon SageMaker
- Examples of fine-tuning Llama 3 8B on a single GPU using 🤗 TRL
Demo: https://huggingface.co/chat/models/meta-llama/Meta-Llama-3-70B-instruct