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 9·5 min read
Introducing North Mini Code: Cohere’s First Model For Developers
TL;DR
Software engineers face challenges in complex coding tasks due to limitations in existing models. Cohere developed North Mini Code, a 30B-parameter Mixture-of-Experts model optimized for agentic coding tasks.
✦ Why It Matters
Engineers can leverage North Mini Code for improved performance in complex coding tasks and software development workflows.
Key Takeaways
- Explore North Mini Code for advanced coding tasks.
- Utilize its Mixture-of-Experts architecture for better performance.
- Leverage its training on real-world coding environments.
- Access the model via Hugging Face and Cohere API.
How It Works
North Mini Code employs a Mixture-of-Experts architecture, activating a subset of 128 experts per token to optimize performance. It uses a combination of sliding-window and global attention mechanisms to efficiently process complex coding tasks.
The model undergoes a two-stage training process, first focusing on diverse coding capabilities and then refining its skills through reinforcement learning with verifiable rewards.