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
Hacker News·21d ago·5 min read
Bonsai 27B (1-bit LLM): The First 27B-Class Model to Run on a Phone
TL;DR
Bonsai 27B is the first 27 billion parameter language model capable of running on a phone, specifically the iPhone 17 Pro. It utilizes 1-bit and ternary weight representations to drastically reduce memory requirements while maintaining advanced capabilities like multi-step reasoning and multimodal processing.
✦ Why It Matters
Engineers can now develop and deploy advanced AI applications directly on mobile devices using Bonsai 27B.
Key Takeaways
- Experiment with Bonsai 27B for mobile AI applications.
- Utilize low-bit models to reduce memory requirements.
- Leverage local execution for enhanced privacy.
- Explore hybrid deployments for cost-effective AI solutions.
Full Summary
Bonsai 27B, based on the Qwen3.6 architecture, is a groundbreaking multimodal language model that can operate on mobile devices. Traditional 27B models require excessive memory, making them impractical for phones.
Bonsai 27B introduces two variants: Ternary Bonsai 27B at 5.9 GB for laptops, and 1-bit Bonsai 27B at 3.9 GB for phones, both maintaining impressive performance metrics. In benchmark tests, Ternary Bonsai 27B retains 95% of full-precision performance, while 1-bit Bonsai 27B retains 90%.
This model supports complex tasks like structured tool calls and vision processing, with a context length of 262K tokens. The ability to run locally on devices allows for persistent, offline AI agents, enhancing privacy and reducing operational costs.