CUDA: build archs as virtual for GGML_NATIVE=OFF

Author: JohannesGaessler

ggml/src/ggml-cuda/CMakeLists.txt

@@ -15,9 +15,9 @@ if (CUDAToolkit_FOUND)
         if (GGML_NATIVE AND CUDAToolkit_VERSION VERSION_GREATER_EQUAL "11.6" AND CMAKE_VERSION VERSION_GREATER_EQUAL "3.24")
             set(CMAKE_CUDA_ARCHITECTURES "native")
         elseif(GGML_CUDA_F16 OR GGML_CUDA_DMMV_F16)
-            set(CMAKE_CUDA_ARCHITECTURES "60;61;70;75;80")
+            set(CMAKE_CUDA_ARCHITECTURES "60-virtual;61-virtual;70-virtual;75-virtual;80-virtual")
         else()
-            set(CMAKE_CUDA_ARCHITECTURES "50;61;70;75;80")
+            set(CMAKE_CUDA_ARCHITECTURES "50-virtual;61-virtual;70-virtual;75-virtual;80-virtual")
         endif()
     endif()
     message(STATUS "Using CUDA architectures: ${CMAKE_CUDA_ARCHITECTURES}")