Gemma 3 270M
#14 in Lütte & Edge-Modellegoogle · v3 · 270m · siet 2025-08-14 · 3× · tolest 29. Juni 2026
14
Momentum
Gemma 3 270M is the smallest model in Google DeepMind's Gemma 3 family, released on August 14, 2025. It is a text-only model with 270 million parameters (170M embedding parameters and 100M transformer block parameters), trained on 6 trillion tokens. The model is designed for task-specific fine-tuning (e.g., classification, entity extraction, query routing) and explicitly targets edge devices such as smartphones and IoT hardware. As an open-weights model it is freely available and can run entirely offline on-device; the INT4-quantized version requires only approximately 125 MB of RAM.
Momentum-Verloop
04.04.03.07.
Features
| Context Window | 32,768 tokens (32K) |
| Model Size (Parameters) | 270 million parameters total (of which 170M embedding parameters, 100M transformer block parameters) |
| Price Tier | Free – open weights, $0.00 per 1M input/output tokens; download via Hugging Face, Kaggle, Ollama, LM Studio, among others, under the Gemma Terms of Use (commercial use allowed with restrictions) |
| Memory Footprint (GB) | approx. 0.5 GB (BF16/full precision); approx. 125 MB with INT4 quantization (QAT) |