Llama Guard 3 8B, released by Meta-Llama on February 12, 2025, is a fine-tuned model designed for content safety classification. It uses a large context window of 131072 tokens to analyze and classify text prompts and responses, ensuring they meet safety standards. This model operates in a text-to-text modality, making it versatile for various applications. It supports content moderation in eight languages, enhancing the safety and security of search and code interpreter tool calls. With features like frequency penalty, logit bias, and temperature control, Llama Guard 3 8B provides robust content moderation aligned with MLCommons standards.
Use Cases
Here are a few ways teams apply Llama Guard 3 8B in practice—from fast drafting to multimodal understanding. Adapt these ideas to your workflow.
Moderate content in multilingual platforms
Enhance safety in search tool responses
Secure code interpreter outputs
Classify unsafe content categories
Support large-scale text analysis
Key Features
A quick look at the capabilities that make this model useful in real projects.
Pretrained Llama-3.1-8B model
Classifies content for safety
Supports eight languages
Optimized for search and code tools
Large 131072 token context window
Aligned with MLCommons standards
Specs
Overview
Vendor
meta-llama
Model ID
meta-llama/llama-guard-3-8b
Release
2025-02-12
Modalities & context
Input
text
Output
text
Context
131,072 tokens
Parameters & defaults
Supported parameters: frequency_penalty, logit_bias, max_tokens, min_p, presence_penalty, repetition_penalty, response_format, seed, stop, temperature, top_k, top_p
Defaults: temperature 0.2, top_p 0.95
Benchmark tests: Llama Guard 3 8B
We ran this model against a few representative prompts to show its range. Review the outputs below and be the judge.
Text
Prompt:
Write 150 words on how AI might positively upend work, leisure and creativity
, as well as the broader age old question of whether some new AI creations might be considered art if they could not be interacted with directly.
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
safe
unsafe
S4
Run this prompt on Upend.AI