whisper.cpp / ggml-cuda /fattn-vec-f32.cuh
JohannesGaessler's picture
CUDA: add FP32 FlashAttention vector kernel (llama/7188)
03d4b22 unverified
raw
history blame
114 Bytes
#include "common.cuh"
void ggml_cuda_flash_attn_ext_vec_f32(ggml_backend_cuda_context & ctx, ggml_tensor * dst);