llama.cpp

models: remove redundant ggml_cont calls now supported by CUDA unary kernels

#30198

Merged