Whisper.cpp prebuilt binaries

Whisper.cpp prebuilt binaries for static and dynamic linking

Building on Windows

You will need MSVS 2022 installed.

Set up environment variables, e.g.:

> $env:BUILD_WITH_ACCEL="nvidia"
> $env:CUDA_TOOLKIT_ROOT_DIR="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.5"

Acceleration options: generic, nvidia and amd

For amd make sure $env:HIP_PATH points to the HIP installation folder, e.g. where $env:HIP_PATH\bin\clang.exe would be located. See https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md#hip for more details

The HIP installer can be downloaded from https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q3-Win10-Win11-For-HIP.exe.

You'll also need make which is most easily obtained by installing Chocolatey and then running choco install make

Run the build script:

> ./Build-Windows.ps1 -Version 0.0.11

Building on Mac OS

Set the MACOS_ARCH env variable to x86_64 or arm64:

$ export MACOS_ARCH=x86_64

Set the METAL_STD env variable to 2.4 (for MacOS 12.0+), 3.0 (for MacOS 13.0+), 3.1 (for MacOS 14.0+), or 3.2 (for MacOS 15.0+)

$ export METAL_STD=2.4

To disable CoreML support, set the USE_COREML env var to off

$ export USE_COREML=off

Run the build script:

$ ./build-macos.sh 0.0.11

Building on Linux

Set up environment variables, e.g.:

> $env:BUILD_WITH_ACCEL="nvidia"
> $env:CUDA_TOOLKIT_ROOT_DIR="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.5"

Acceleration options: generic, nvidia and amd

For amd install the ROCm SDK v6.4.2 For nvidia install the CUDA SDK v12.8.0

For all variants install the Vulkan and OpenCL SDKs, as well as OpenBLAS

Run the build script:

$ ./build-linux.sh 0.0.11