Search Results for "swift"

TurboFieldfare is a custom Swift and Metal runtime for running the instruction-tuned Gemma 4 26B-A4B model on Apple Silicon Macs with limited memory. Instead of loading the entire model, it keeps the shared core and KV cache in RAM while streaming only the routed experts required for each token from SSD. This approach reduces active memory use to roughly 2 GB while the installed model occupies about 14.3 GB of storage.

...It provides export recipes that convert supported open-source models into Core AI model files. It also includes Python primitives for authoring custom PyTorch models that are better suited for Apple platform deployment. The Swift package adds runtime utilities that help developers integrate exported models into macOS and iOS apps. The repository also contains agent skills that guide coding assistants through Core AI workflows, model authoring, and compression exploration. It is useful for developers who want a curated path from model preparation to local app integration on Apple hardware.