llama.cpp

fix tensor stride for quantized types in create_tensor

#21397

Open