ggml: avoid creating CUDA context during device init#20595

am17an merged 1 commit into

ggml-org:masterggml-org/llama.cpp:masterfrom

ServeurpersoCom:cuda-defer-context-initServeurpersoCom/llama.cpp:cuda-defer-context-initCopy head branch name to clipboard