llama.cpp

ggml: Re-enable CUDA graphs in presence of CONT and DUP nodes

#12970

Merged