Closed aldorof closed 5 months ago
The same way it's realised on llama.cpp. So that one can build CUDA BLAS for HIP with something like this:
CXX=hipcc CC=hipcc cmake -DWHISPER_HIPBLAS=ON -DAMDGPU_TARGETS="gfx1100;gfx1101;gfx1102" -DBUILD_SHARED_LIBS=ON ../whisper.cpp
The same way it's realised on llama.cpp. So that one can build CUDA BLAS for HIP with something like this: