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
DeepMind Blog·Oct 23·5 min read
Introducing Gemma 3 270M: The compact model for hyper-efficient AI
TL;DR
AI models often require significant resources, making them inefficient for specific tasks. DeepMind has introduced Gemma 3 270M, a compact model designed for task-specific fine-tuning with strong instruction-following capabilities.
✦ Why It Matters
Engineers can leverage Gemma 3 270M for efficient, cost-effective AI solutions tailored to specific tasks.
Key Takeaways
- Utilize Gemma 3 270M for high-volume, well-defined tasks.
- Leverage low power consumption for on-device applications.
- Rapidly fine-tune the model for specific use cases.
- Ensure user privacy by running models entirely on-device.
Full Summary
Gemma 3 270M is the latest addition to the Gemma family, featuring a compact architecture with 270 million parameters, including 170 million for embeddings and 100 million for transformer blocks. This model excels in instruction-following tasks, achieving a new benchmark in performance for its size, as evidenced by the IFEval benchmark.
It is designed for extreme energy efficiency, consuming only 0.75% of battery for 25 conversations on a Pixel 9 Pro SoC. The model is pre-trained and comes with quantization-aware trained (QAT) checkpoints, allowing deployment on resource-constrained devices without significant performance loss.
Developers can fine-tune this model for specific tasks like sentiment analysis or creative writing, enabling rapid deployment and cost-effective solutions. Its ability to run on-device ensures user privacy, making it suitable for applications handling sensitive information.