Re: llama.cpp, whisper.cpp, ggml: Next steps

[Christian Kastner <ckk@debian.org>]

On 2025-02-26 21:29, Petter Reinholdtsen wrote:
> Note, to get the programs in the whisper.cpp package working, one need
> to se LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu/ggml/ or similar.  I
> have not been able to get cmake to install a rpath/runpath into the
> binaries.

Adding the following to the "configure" step worked for me:

    -DCMAKE_INSTALL_RPATH=/usr/lib/$(DEB_TARGET_MULTIARCH)/ggml
    -DCMAKE_INSTALL_RPATH_USE_LINK_PATH=OFF \

    $ ldd debian/libwhisper1/usr/lib/x86_64-linux-gnu/libwhisper.so.1
          linux-vdso.so.1 (0x00007fb42c970000)
          libggml.so => /usr/lib/x86_64-linux-gnu/ggml/glibc-hwcaps/x86-64-v4/libggml.so (0x00007fb42c8df000)
          libggml-base.so => /usr/lib/x86_64-linux-gnu/ggml/glibc-hwcaps/x86-64-v4/libggml-base.so (0x00007fb42c830000)

I didn't do any testing beyond that, though. Things have come up, and I
won't be able to do much more until Saturday, sadly.

By the way -- why TARGET instead of HOST? Genuinely curious, I thought
its use case was just for toolchains.

Best,
Christian