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)

Open-weights safety classifier under Meta's Llama Community License, which carries use restrictions and is not OSI-approved, so it lands at the open_weights tier (3) like the rest of the Llama family rather than open_source.

• https://huggingface.co/meta-llama/Llama-Guard-3-8B recorded 2026-08-13

  Repo carries license - llama3.1 and four safetensors shards, gated manual, so the weights are downloadable after accepting Meta's terms. The card has a Training Data section but publishes no mixture and links no dataset, and cardData declares no datasets. It points readers at Llama Recipes on Meta's GitHub for configuration, which is the inference recipe rather than a training pipeline.

• https://huggingface.co/api/models/meta-llama/Llama-Guard-3-8B recorded 2026-08-13

  tags include license:llama3.1; gated - manual; private - false; siblings list model-00001-of-00004.safetensors through model-00004-of-00004.safetensors; cardData carries no datasets key.

164,619 downloads in the trailing 30 days for the one declared artifact, meta-llama/Llama-Guard-3-8B, which bands at 100K-1M, level 3 on the software and model adoption scale.

• https://huggingface.co/api/models/meta-llama/Llama-Guard-3-8B recorded 2026-08-13

  164,619 downloads in the trailing 30 days for meta-llama/Llama-Guard-3-8B

Broad hazard taxonomy and multilingual coverage; the reference open guardrail others are measured against, and several bands in this category are recorded relative to it.

• https://huggingface.co/meta-llama/Llama-Guard-3-8B recorded 2026-08-13

  Chat template lists the unsafe content categories S1 Violent Crimes through S14 Code Interpreter Abuse - fourteen. Card states it 'was aligned to safeguard against the MLCommons standardized hazards taxonomy' and 'provides content moderation in 8 languages'.