Llama Guard

weights

open(Llama-Guard-3-8B safetensors on HF)

data

not-released

code

inference recipe public

license

Llama-3.1-Community(NOT OSI, >700M-MAU restriction)

Meta publishes the classifier's weights under the Llama 3.1 Community License, which carries use restrictions and is not OSI-approved. The training data is described on the model card but not released, and only the inference recipes are public.

• https://huggingface.co/api/models/meta-llama/Llama-Guard-3-8B recorded 2026-09-29

  The Hugging Face API record for the 8B model carries the tag "license:llama3.1", is gated for manual approval, and lists the weights as "model-00001-of-00004.safetensors" through "model-00004-of-00004.safetensors"; its cardData declares no datasets.

• https://huggingface.co/meta-llama/Llama-Guard-3-8B recorded 2026-09-29

  The model card's Training Data section describes the data only in prose ("In order to scale training data for new categories and new capabilities such as multilingual and tool use") and links no dataset, and it points to "Llama Recipes" for configuration; the embedded license requires licensees above "700 million monthly active users in the preceding calendar month, you must request a license from Meta".

Adoption is measured by Hugging Face downloads for the one artifact Meta has declared, meta-llama/Llama-Guard-3-8B. No other channel is declared for the model, so this single download count is the only visible measure of its use.

• https://huggingface.co/api/models/meta-llama/Llama-Guard-3-8B recorded 2026-10-07

  43,382 downloads in the trailing 30 days for meta-llama/Llama-Guard-3-8B

Llama Guard classifies both prompts and responses against a fourteen-category hazard taxonomy aligned to MLCommons, in eight languages, and also screens search and code-interpreter tool calls.

• https://huggingface.co/meta-llama/Llama-Guard-3-8B recorded 2026-09-29

  The model card's chat template lists the categories from "S1: Violent Crimes" to "S14: Code Interpreter Abuse", and the card says the model "was aligned to safeguard against the MLCommons standardized hazards taxonomy", "provides content moderation in 8 languages" and "was optimized to support safety and security for search and code interpreter tool calls".