llama.cpp
CUDA: fix scratch buffers being allocated on non-main device
#3220
Merged

Loading