Llama Guard 2
Meta's earlier content-safety classifier model, preceding Llama Guard 3
Llama Guard 2 is a content-safety classifier Meta released alongside Llama 3 in April 2024, built on an 8B Llama 3 base rather than as a general-purpose chat model. It checks both prompts going into a language model and responses coming out, flagging content across 11 hazard categories drawn from the MLCommons taxonomy, covering things like violence, hate speech, and sexual content. Developers run it as a guardrail in front of or behind another LLM in a pipeline, rather than using it to generate answers directly. Meta has since replaced it with Llama Guard 3, but it remains available on Hugging Face under the Llama 3 Community License for teams already running it in production.