TrainingSafety
Llama 2: Open Foundation and Fine-Tuned Chat Models
Meta·July 18, 2023
Hugo Touvron, Louis Martin, Kevin Stone
View on arXivTL;DR
Releases Llama 2 with open weights and a detailed account of its supervised fine-tuning and RLHF chat alignment.
Why it matters
The release that catalyzed the open-weights ecosystem, giving the community a strong, commercially usable base and a transparent alignment recipe.