llama.cpp

cuda, sycl : fix batched gemm when ne02 == 1 && ne03 > 1

#15038

Merged