EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Jul 29·5 min read
Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac
TL;DR
TurboFieldfare is an open-source engine that enables the Gemma 4 26B model to run on Apple Silicon Macs with as little as 2 GB of RAM. It achieves this by streaming only necessary model components from SSD, making it feasible for devices with limited memory.
✦ Why It Matters
Try TurboFieldfare today to run large AI models on your Apple Silicon Mac with limited RAM.
Key Takeaways
- Clone the TurboFieldfare repository to get started.
- Use the command-line interface for advanced model interactions.
- Benchmark performance on different Apple Silicon models.
- Explore the community guide for contributing performance results.
How It Works
TurboFieldfare operates by maintaining a small core of the model in memory while streaming additional components from SSD as needed. This allows it to efficiently manage memory usage, making it possible to run a large model on devices with limited RAM.
The system uses a combination of Swift and Metal to optimize performance, leveraging custom kernels for various operations.