Gemma 3 270M
Gemma 3 270M: tiny model (170M embedding, 100M transformer parameters) for task-specific fine-tuning; INT4 used 0.75% of a Pixel 9 Pro battery over 25 chats.
Pushes Gemma to the very small end with a 256k-token vocabulary, instruction-tuned and pretrained checkpoints, and quantization-aware-trained INT4 weights. Gemma downloads passed 200 million the week before.
- Date
- Thursday, 14 August 2025
- Lab
- Google DeepMind
- Kind
- open-weights
- Access
- open weights (restricted license)
Sources
This record was checked against its sources on 6 October 2026. How we check