● Covered by 1 source · 1 reportMedium impact1 neutral
Open-source engine runs Gemma 4 26B on Apple Silicon Macs with 2GB RAM
Key points
- Runs Gemma 4 26B-A4B with ~2 GB RAM on Apple Silicon Macs.
- Streams model experts from SSD to reduce memory footprint.
- Written in Swift and Metal, includes CLI and Mac app.
- Supports 8 GB M2 MacBook Air and macOS 26 with Metal 4.
Efficient LLM Inference on Apple Silicon
TurboFieldfare is an open-source engine designed to run the Gemma 4 26B-A4B large language model on Apple Silicon Macs. It achieves this with a memory footprint of approximately 2 GB, significantly less than the model's full 14.3 GB size. This efficiency allows the 26-billion-parameter model to operate on Macs with as little as 8 GB of unified memory.
Technical Implementation
The engine's core innovation lies in its memory management strategy. Instead of loading the entire model into RAM, TurboFieldfare keeps a 1.35 GB core and FP16 KV cache in memory. It then streams only the necessary 'experts' from the SSD as each token is processed. The runtime, streaming installer, command-line interface (CLI), and native Mac application are all developed using Swift and Metal.
User Experience and Requirements
Users can install TurboFieldfare by cloning its GitHub repository and building it with Swift. The first run involves downloading and repacking the 15 GB model. The software provides a native Mac app, a CLI, and an experimental OpenAI-compatible server. It requires an Apple Silicon Mac (validated on an 8 GB M2 MacBook Air), macOS 26 with Metal 4, Xcode 26 or newer, and sufficient storage for the model installation.
Impact and Future Development
This development makes large language models like Gemma 4 26B more accessible to users with consumer-grade Apple hardware, bypassing the need for extensive RAM upgrades. The project emphasizes community contributions for benchmarking, acknowledging that factors like prompt length, generated length, and hardware configurations influence throughput. The curated experiment record includes 103 measured results across various performance aspects.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub cmc_internal/api GitHub github/collect GitHub drumih/turbo-fieldfare GitHub _private/browser GitHub get-started/accessibility GitHub open-source/sponsorsReporting from
TurboFieldfare, an open-source engine, enables the Gemma 4 26B-A4B model to run on Apple Silicon Macs using approximately 2 GB of RAM. This is achieved by streaming experts from SSD as needed, allowing the 26-billion-parameter model to operate on devices with limited memory, such as 8 GB M-series Macs.