llama.cpp

0cbee131 - cuda/vulkan: specify fp32-only support for some operations in supports_op (ggml/1129)