We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hugging Face Blog·Jun 4·5 min read
Nemotron 3.5 Content Safety: Customizable Multimodal Safety for Global Enterprise AI
TL;DR
A gap existed in multimodal safety evaluations, where interactions between text and images could lead to overlooked policy violations. Nemotron 3.5 was developed to evaluate user prompts, images, and assistant responses together, providing a unified safety verdict.
✦ Why It Matters
Engineers can implement Nemotron 3.5 to enhance safety evaluations in their AI applications across various industries.
Key Takeaways
- Utilize Nemotron 3.5 for multimodal safety evaluations.
- Leverage custom policy enforcement for industry-specific applications.
- Enable THINK mode for auditable reasoning in safety decisions.
- Deploy across 140 languages without separate fine-tuning.
How It Works
Nemotron 3.5 processes user prompts, images, and assistant responses together, allowing it to identify safety violations that arise from their interactions. It uses a 4B-parameter architecture with a 128K context window, enabling efficient reasoning and classification across multiple languages and modalities.