llama.cpp

85f99dca - ggml: support concat for scalar types at cuda backend (#24011)