The Gemma 3 270M is a highly compact, 270-million parameter multimodal model from Google designed for extreme efficiency. It is specifically engineered for well-defined tasks like text classification and data extraction where it offers a cost-effective alternative to larger models. Its lightweight architecture is particularly optimized for rapid fine-tuning.

Gemma3 270M been optimized for the Synaptics Astra™ SL2610-Series processors with Torq NPU.

The compiled model files are available for download on Huggingface at Synaptics/Gemma-3-270m-it.

Usage tutorial to be available in the future at Synaptics AI Developer Zone.

Both the source model and the compiled model for on-device deployment are licensed under MIT License.