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·15d ago·5 min read
Introducing Cosmos 3 Edge
TL;DR
NVIDIA has released Cosmos 3 Edge, a 4-billion-parameter open world model designed for edge devices. It enables robots and vision AI agents to understand their environments and make real-time decisions.
✦ Why It Matters
Engineers can implement Cosmos 3 Edge to improve real-time decision-making in robotics applications today.
Key Takeaways
- Utilize Cosmos 3 Edge for efficient edge AI applications.
- Fine-tune the model for specific robotic tasks using provided scripts.
- Leverage the dual transformer architecture for enhanced reasoning.
- Explore the open Cosmos Frameworks for custom model adaptations.
How It Works
Cosmos 3 Edge integrates two transformer towers: an autoregressive tower for processing vision and text tokens, and a diffusion tower for handling vision, audio, and action tokens. This design allows the model to reason about scenes and generate outputs based on a shared multimodal representation, enabling it to predict the consequences of actions and infer actions from observed effects.