cuda : fix FA Q src index (1 -> 0) (llama/9374)

diff --git a/src/ggml-cuda/fattn.cu b/src/ggml-cuda/fattn.cu
index f87f33b3e574b65af5a8ceaff958fc98c7c69c57..f28a19d40b35617d34ff42c126c1d8bce2010433 100644 (file)
--- a/src/ggml-cuda/fattn.cu
+++ b/src/ggml-cuda/fattn.cu
@@ -152,7 +152,7 @@ static void ggml_cuda_flash_attn_ext_wmma_f16(ggml_backend_cuda_context & ctx, g
     }                                                                       \
 
 static void ggml_cuda_flash_attn_ext_vec_f16(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
-    ggml_tensor * Q = dst->src[1];
+    ggml_tensor * Q = dst->src[0];
     ggml_tensor * K = dst->src[1];
     ggml_tensor * V = dst->src[2];
 
@@ -227,7 +227,7 @@ static void ggml_cuda_flash_attn_ext_vec_f16(ggml_backend_cuda_context & ctx, gg
     }                                                                       \
 
 static void ggml_cuda_flash_attn_ext_vec_f32(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
-    ggml_tensor * Q = dst->src[1];
+    ggml_tensor * Q = dst->src[0];
     ggml_tensor * K = dst->src[1];
     ggml_tensor * V = dst->src[2];