drumih/turbo-fieldfare

Why this repo matters

• •Enables Gemma 4 26B inference in 2GB RAM on Apple Silicon MacBooks using Metal.
• •Demonstrates efficient on-device LLM deployment via native Swift and GPU acceleration.
• •Provides accessible local AI reference implementation for resource-constrained macOS environments.