llama.cpp

4de4f8fc - llama + CUDA: flash_attn_ext_rows (fix unified kv for multi-seq)