llama.cpp

166e60bf - ggml : ggml_flash_attn_ext() support ALiBi (CPU)