# [Let's Paper Party!] Llama 2: Open Foundation and Fine-Tuned Chat Models

Arxiv Link: [llama 2: open foundation and fine-tuned chat models](https://arxiv.org/abs/2307.09288)

Here's the keynote of this paper

* Llama 2 Release: The paper discusses the release of Llama 2, an open-source large language model (LLM) that is free for research and commercial use1. It is designed to enable experimentation, innovation, and responsible scaling of ideas.
* Model Accessibility: Llama 2 is available for download, including model code, weights, a user guide, a responsible use guide, a license, an acceptable use policy, and a model card with technical specifications.
* Training and Benchmarks: The model was pretrained on 2 trillion tokens and offers double the context length of its predecessor, Llama 12. Llama Chat, a fine-tuned version, includes over 1 million human annotations34.
* Safety and Helpfulness: Llama Chat employs reinforcement learning from human feedback (RLHF) to ensure the model’s safety and helpfulness, using techniques like rejection sampling and proximal policy optimization (PPO)56.
* Open Innovation and Responsibility: The paper emphasizes Meta’s commitment to responsible AI development, offering resources like a Responsible Use Guide and Safety Red-teaming for developers and researchers.