Fix Cuda offloading in llava#3621

KerfuffleV2 merged 1 commit intomasterggml-org/llama.cpp:masterfrom

llava-fix-offloadingggml-org/llama.cpp:llava-fix-offloadingCopy head branch name to clipboard