It turns out that this is a scikit-build issue:
https://github.com/scikit-build/scikit-build/issues/740
https://github.com/scikit-build/scikit-build/issues/939
https://github.com/scikit-build/scikit-build/issues/981
--
https://mail.python.org/mailman/listinfo/python-list
On Wed, Jun 21, 2023 at 5:45 PM Cebtenzzre wrote:
> In the case of bitsandbytes, there is no build folder generated at
> all, just bitsandbytes.egg-info. This results in an error when I try
> to import it, e.g. "CUDA SETUP: Required library version not found:
> libbitsandbytes_cuda121_nocublaslt.s
When I try to use `pip install -e .` with llama-cpp-python or
bitsandbytes, it doesn't build the shared libraries like `pip install
.` does.
In the case of llama-cpp-python, there is no _skbuild folder generated
at all, just llama_cpp_python.egg-info. This results in an error when
I try to import