LlavaGuard-v1.2-0.5B-OV-hf
| Property | Value |
|---|---|
| Parameter Count | 894M |
| Model Type | Image-Text-to-Text |
| Architecture | LLaVA-OneVision Based |
| Paper | LlavaGuard-Arxiv |
| Tensor Type | FP16 |
What is LlavaGuard-v1.2-0.5B-OV-hf?
LlavaGuard-v1.2-0.5B-OV-hf is a specialized vision-language model designed for content moderation and safety assessment. Built upon the LLaVA-OneVision 0.5B architecture, it represents the smallest variant in the LlavaGuard family while maintaining robust performance in safety analysis capabilities. The model is specifically trained on the LlavaGuard-DS dataset and features a substantial 32K token context window.
Implementation Details
The model implements a comprehensive safety taxonomy covering nine distinct categories, from hate speech to disaster content. It utilizes the Transformers framework for inference and provides detailed safety assessments through a structured JSON output format.
- Built on llava-onevision-qwen2-0.5b-ov base model
- Supports HF Transformers inference pipeline
- Implements FP16 precision for efficient processing
- Features extensive safety policy categorization system
Core Capabilities
- Visual content safety assessment across 9 categories
- Structured safety rating output (Safe/Unsafe)
- Detailed rationale generation for safety decisions
- Support for image-text conversation format
- Efficient processing with 894M parameter footprint
Frequently Asked Questions
Q: What makes this model unique?
LlavaGuard stands out for its specialized focus on safety assessment and content moderation, combining visual and textual understanding with a comprehensive safety taxonomy. Its efficient 0.5B parameter size makes it practical for deployment while maintaining strong performance.
Q: What are the recommended use cases?
The model is primarily targeted toward researchers and is designed for research applications in content moderation, safety assessment of visual content, and development of safer AI systems. It's particularly useful for analyzing images against a detailed safety policy framework.
