useGpu property

bool useGpu

Whether to use hardware-accelerated inference.

On Android, this requests the LiteRT GPU accelerator, with automatic fallback to CPU when the model can't compile on the GPU. On iOS, this enables hardware-accelerated inference: Core ML on the Neural Engine + CPU (iOS 16+), or an opt-in Core AI .aimodel across the Neural Engine, GPU and CPU (iOS 27+); when false, inference is pinned to CPU only.

Default is true for better performance, but can be set to false for stability on devices where accelerated inference causes crashes.

Implementation

final bool useGpu;