windows build does not work

[johnny-smitherson]

The windows build instruction mentions this:

https://github.com/ShelbyJenkins/llm_client/blob/master/docs/build.md

Windows

Requires make see the suggested make instructions here or if using WSL you can run the dev container in this repo.

But the instructions linked at https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md do not actually mention make in any way.

I looked at the blame, and the docs have been removed for make 3 months ago -

• new version: https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md
• old version: https://github.com/ggml-org/llama.cpp/blob/64ed2091b24b2f9747148fdf49a34ed5938762c3/examples/server/server.cpp
• diff / pr : "make" was deprecated: make : deprecate ggml-org/llama.cpp#10514

The "make" workflow seems to have some problems:

• does not work if cuda is installed in a path with spaces (such as all windows installations that go in "C:/program fiels")
• after that is fixed, still does not work, one gets all the errors mentioned here: Error on make LLAMA_CUBLAS=1 ggml-org/llama.cpp#1470
• solution in that other thread was "use cmake"

It seems the windows builds are broken until the build script is changed to use "cmake" with a new version of llamacpp

[ljt019]

+1, was looking forward to trying this crate out, guess it's back to llama-cpp-2

[ShelbyJenkins]

@johnny-smitherson and @ljt019 the new method builds for windows using cmake (or falls back to downloading the binary from the release.) I got it to build on windows, but it was convoluted and required installing all the dependencies listed in the llama.cpp windows build docs.