app-accessibility/whisper-cpp
Port of OpenAI's Whisper model in C/C++
-
whisper-cpp-9999blas cuda ffmpeg hip opencl sdl2 vulkan
View
Download
Browse License: MIT Overlay: stuff -
whisper-cpp-9999blas cuda ffmpeg hip opencl sdl2 vulkan
View
Download
Browse License: MIT Overlay: guru -
whisper-cpp-1.8.5~amd64blas cuda ffmpeg hip opencl sdl2 vulkan
View
Download
Browse License: MIT Overlay: stuff -
whisper-cpp-1.8.4~amd64blas cuda ffmpeg hip opencl sdl2 vulkan
View
Download
Browse License: MIT Overlay: guru -
whisper-cpp-1.8.4~amd64blas cuda ffmpeg hip opencl sdl2 vulkan
View
Download
Browse License: MIT Overlay: stuff -
whisper-cpp-1.8.3-r2~amd64blas cuda hip opencl sdl2 vulkan
View
Download
Browse License: MIT Overlay: guru -
whisper-cpp-1.7.6~amd64blas cuda hip opencl sdl2 vulkan
View
Download
Browse License: MIT Overlay: guru -
whisper-cpp-1.5.4~amd64blas cuda examples hip +models opencl sdl2 test
View
Download
Browse License: MIT Overlay: jm-overlay -
whisper-cpp-1.4.2~amd64 ~ppc64debug +command +models +quantize +stream
View
Download
Browse License: MIT Overlay: luke-jr -
whisper-cpp-1.2.1-r1~amd64blas examples +models sdl2 test
View
Download
Browse License: MIT Overlay: jm-overlay
ChangeLog
commit 8d2e44f46002a92f6456b7a13147ea0970730bf6
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri May 29 13:43:59 2026 +0200
app-accessibility/whisper-cpp: add 1.8.5
commit 22414d8cbda8cd689705199244ab2eb5e31353f3
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 14 17:01:19 2026 +0200
app-accessibility/whisper-cpp: new package, add 1.8.4 and 9999
Forked from guru; fixes vs. upstream:
- -DGGML_CUBLAS (silently ignored) → -DGGML_CUDA (verified 2026-05-14)
- BDEPEND shaderc gated on vulkan? (glslc only needed in ggml-vulkan backend)
- -DGGML_NATIVE=OFF to prevent -march=native injection; with
GGML_NATIVE_DEFAULT=ON and GGML_NATIVE=OFF, ggml's hand-optimised
AVX2/FMA/F16C kernels compile via INS_ENB=ON
- -DGGML_CCACHE=OFF in the sandbox
- -DWHISPER_BUILD_TESTS=OFF (tests require model files not in tarball)
- CMAKE_CUDA_HOST_COMPILER pinned to g++-15 when cuda USE is set
(CUDA 13.x nvcc rejects gcc>15; verified 2026-05-14)
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri May 29 13:43:59 2026 +0200
app-accessibility/whisper-cpp: add 1.8.5
commit 22414d8cbda8cd689705199244ab2eb5e31353f3
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 14 17:01:19 2026 +0200
app-accessibility/whisper-cpp: new package, add 1.8.4 and 9999
Forked from guru; fixes vs. upstream:
- -DGGML_CUBLAS (silently ignored) → -DGGML_CUDA (verified 2026-05-14)
- BDEPEND shaderc gated on vulkan? (glslc only needed in ggml-vulkan backend)
- -DGGML_NATIVE=OFF to prevent -march=native injection; with
GGML_NATIVE_DEFAULT=ON and GGML_NATIVE=OFF, ggml's hand-optimised
AVX2/FMA/F16C kernels compile via INS_ENB=ON
- -DGGML_CCACHE=OFF in the sandbox
- -DWHISPER_BUILD_TESTS=OFF (tests require model files not in tarball)
- CMAKE_CUDA_HOST_COMPILER pinned to g++-15 when cuda USE is set
(CUDA 13.x nvcc rejects gcc>15; verified 2026-05-14)

