L

Llama Guard 3 11B Vision

Developed by meta-llama
A multimodal content security classifier fine-tuned based on Llama-3.2-11B, optimized for detecting harmful text-image hybrid content
Downloads 4,553
Release Time : 9/20/2024

Model Overview

Llama Guard 3 Vision is a classifier designed to protect the input (prompt classification) and output (response classification) content safety of large language models, particularly excelling in handling multimodal content that includes images.

Model Features

Multimodal Content Security Detection
Specially optimized for detecting harmful text+image combinations, with enhanced capability to identify risks in image-based character recognition
Custom Classification Rules
Supports user-defined risk classification systems or exclusion of default categories, providing flexible configuration options
Low False Positive Rate
Response classification false positive rate is only 0.016%, significantly outperforming similar models

Model Capabilities

Multimodal Content Analysis
Harmful Content Identification
Risk Classification
Image Understanding
Text Generation

Use Cases

Content Moderation
Social Media Content Moderation
Automatically detects whether user-uploaded text-image content contains prohibited information such as violence or hate speech
F1 score reaches 0.938 (response classification)
AI Security
Large Language Model Protection
Acts as a pre-filter to prevent large language models from processing or generating harmful content
Precision rate of 0.961 (response classification)
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase