whisper.cpp / ggml-cuda /fattn.cuh
ggerganov's picture
ggml : add Flash Attention (llama/5021)
34d3b03
raw
history blame
106 Bytes
#include "common.cuh"
void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst);