AI BriefingMeta Releases Llama 2 and Integrates with HF
·2023.07.18 09:00
Key point
Meta's open-source LLM Llama 2 has been released, and it is provided integrated with models and various tools through Hugging Face.
Details
Meta has unveiled a new open-access large language model (LLM) family, Llama 2. This model offers a permissive license that allows commercial use, and is deployed with tight integration into the Hugging Face ecosystem.
Key Features and Model Configuration
- Model Scale: Pretrained models with 7B, 13B, and 70B parameters, along with the conversation-optimized Llama 2-Chat model.
- Performance Improvements: Trained on 40% more tokens than Llama 1, with a 4k context length and Grouped-Query Attention technology applied for faster inference on the 70B model.
- RLHF Applied: The Llama 2-Chat model maximizes conversational performance and safety through Reinforcement Learning from Human Feedback (RLHF).
Hugging Face Integration Features
- Hugging Face Hub: Provides 12 open-access models (Base and Chat models).
- Fine-tuning: Examples provided for fine-tuning smaller models on a single GPU.
- Inference: Supports efficient production inference through Text Generation Inference (TGI) and Inference Endpoints.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.