llama.cpp

1efae9b7 - llm : prevent from 1-D tensors being GPU split (#3697)