GPU gets used and BLAS parameter is 1 when I download the CUDA supported llamacpp python version from this url (https://jllllll.github.io/llama-cpp-python-cuBLAS-wheels/AVX2/cu117/llama-cpp-python/)

[shaunck96]

Why doesn't GPU get recognized when I download from here instead ? (https://abetlen.github.io/llama-cpp-python/whl/cu121/llama-cpp-python/). My cluster Driver and worker CUDA version is 12.2. Any suggestions to why this is happening as I want to leverage the CUDA 12.1 supported whl file with llamacppversion >0.2.72 to use the CVE fixed version for prod