Gemma 4 Models Compressed 7x for Edge Devices

🤖 Gemma 4 is now 7x more compact for Edge devices

The Stage.AI team introduced the edge-lm project, allowing compressed Gemma 4 checkpoints to run on Apple Silicon via MLX. Thanks to PLE architectures and vector quantization, model sizes have been reduced by 6.4–7x (e.g., E2B: 1.44 GB instead of 9.26 GB) while maintaining high accuracy.

🌍 This is the key to developing local on-device assistants.

👤 Gemma 4 models now run directly on iPhone or Mac with minimal memory consumption.

Source 1: https://github.com/TheStageAI/edge-lm Source 2: https://app.thestage.ai/blog/7x-size-reduction-for-Gemma4-Edge-models?id=14