From cff6645ccb664aeb20e5e906f9b6fd729b01e1f6 Mon Sep 17 00:00:00 2001
From: Oleg Skutte <00.00.oleg.00.00@gmail.com>
Date: Wed, 28 Aug 2024 00:17:26 +0400
Subject: [PATCH] sync: update ggml

---
 ggml            | 2 +-
 ggml_extend.hpp | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/ggml b/ggml
index 21f9e5c42..21d3a308f 160000
--- a/ggml
+++ b/ggml
@@ -1 +1 @@
-Subproject commit 21f9e5c426b105841c2e346d8f1aafec398edf15
+Subproject commit 21d3a308fcb7f31cb9beceaeebad4fb622f3c337
diff --git a/ggml_extend.hpp b/ggml_extend.hpp
index bc920eef6..810f2b9ef 100644
--- a/ggml_extend.hpp
+++ b/ggml_extend.hpp
@@ -741,7 +741,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_nn_attention_ext(struct ggml_context*
         v = ggml_cont(ctx, ggml_permute(ctx, v, 0, 2, 1, 3));  // [N, n_head, L_k, d_head]
         v = ggml_reshape_3d(ctx, v, d_head, L_k, n_head * N);  // [N * n_head, L_k, d_head]
         LOG_DEBUG("k->ne[1] == %d", k->ne[1]);
-        kqv = ggml_flash_attn_ext(ctx, q, k, v, mask, scale, 0);
+        kqv = ggml_flash_attn_ext(ctx, q, k, v, mask, scale, 0, 0);
     } else {
         v = ggml_cont(ctx, ggml_permute(ctx, v, 1, 2, 0, 3));  // [N, n_head, d_head, L_k]
         v = ggml_reshape_3d(ctx, v, L_k, d_head, n_head * N);  // [N * n_head, d_head, L_k]