llama-cpp-python/vendor/llama.cpp/ggml-cuda/fattn.cuh

#include "common.cuh"

void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst);