From 0a6542b503af91058c2f1a676ca38ace17e33395 Mon Sep 17 00:00:00 2001
From: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Date: Mon, 3 Mar 2025 18:52:34 +0200
Subject: [PATCH] CUDA FA WIP - It actually works!

No TG yet, but for PP I can run FA with fp16 cache and it gets
the same answer.
---
 ggml/src/ggml-cuda/fattn.cu | 1 +
 1 file changed, 1 insertion(+)

diff --git a/ggml/src/ggml-cuda/fattn.cu b/ggml/src/ggml-cuda/fattn.cu
index bbfc1ccb..cdfd5161 100644
--- a/ggml/src/ggml-cuda/fattn.cu
+++ b/ggml/src/ggml-cuda/fattn.cu
@@ -343,6 +343,7 @@ static void ggml_cuda_flash_attn_ext_vec_f32(ggml_backend_cuda_context & ctx, gg
 void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
     const ggml_tensor * KQV = dst;
     const ggml_tensor * Q   = dst->src[0];
+    const ggml_tensor * V   = dst->src[2];
 
     ggml_cuda_set_device(ctx.device);
     const int cc = ggml_cuda_info().devices[ggml_cuda_get_device()].cc;