ollama - 💡(How to fix) Fix Gemma4 takes up too much space [2 comments, 3 participants]

TL;DR

The large memory footprint of the E2B model in Ollama needs to be reduced to make it suitable for low-end devices.

Guidance

• Investigate the differences in model implementation between Google's AI Edge Gallery app and Ollama to identify the cause of the large memory footprint.
• Consider optimizing the E2B model in Ollama for lower memory usage, potentially by using model pruning, quantization, or knowledge distillation techniques.
• Evaluate the system requirements for running the E2B model on low-end devices, such as Raspberry Pis, to ensure that the model can run efficiently on these devices.

Notes

The issue may be specific to the implementation of the E2B model in Ollama, and resolving it may require collaboration with the Ollama development team.

Recommendation

Apply workaround: Optimize the E2B model in Ollama for lower memory usage, as the current implementation is not suitable for low-end devices.