The pre-built Maven Central artifacts include CPU-only native binaries everywhere except macOS, where Metal is enabled by default. If you need a GPU backend such as Vulkan or CUDA you must build Llamatik from source and pass the appropriate CMake flags.

Extra CMake flags#

The llamatik.cmake.args Gradle property (or the LLAMATIK_CMAKE_ARGS environment variable as a fallback) lets you append flags to the CMake configure step without editing the build script.

Values are split on whitespace. Flags containing spaces are not supported.

Vulkan (Android / Desktop Linux)#

./gradlew :library:build -Pllamatik.cmake.args="-DGGML_VULKAN=ON"

CUDA (Desktop Linux / Windows)#

./gradlew :library:build -Pllamatik.cmake.args="-DGGML_CUDA=ON"

Using the environment variable#

LLAMATIK_CMAKE_ARGS="-DGGML_VULKAN=ON" ./gradlew :library:build

Multiple flags#

./gradlew :library:build -Pllamatik.cmake.args="-DGGML_VULKAN=ON -DGGML_VULKAN_VALIDATE=ON"

Scope#

The extra flags are forwarded to these targets:

iOS limitation#

The mergeLlamaStatic* step that produces the iOS fat static library merges a hardcoded list of ggml archives. If a GPU backend produces additional archives (for example a Vulkan MoltenVK layer), they will not be included in the merged binary. A build warning is emitted when extra flags are passed to an iOS target as a reminder. Desktop and Android GPU builds are not affected.

Enabling GPU layers at runtime#

Once you have built with a GPU backend enabled, set gpuLayers in GenerateParams to offload computation:

LlamaBridge.updateGenerateParams(
    gpuLayers = -1  // offload all layers
)

See Generation parameters for the full gpuLayers reference.