Open
Description
I'm using CentOS 7 (glibc 2.17) with both CUDA 11.8 and 12.4.
Up until version 0.3.7, I was able to install llama-cpp-python with either CUDA version. However, starting from 0.3.8, it seems that the cu118 variant is no longer supported, and the package installs in CPU mode instead.
Has anyone managed to get it working with a similar setup?
Here's the install command I'm using:
FORCE_CMAKE=1 CMAKE_ARGS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES='70;75'" pip install llama-cpp-python==0.3.8
Metadata
Metadata
Assignees
Labels
No labels