diff --git a/.SRCINFO b/.SRCINFO
index 9abcf2288e411..018f593f215fe 100644
--- a/.SRCINFO
+++ b/.SRCINFO
@@ -1,6 +1,6 @@
 pkgbase = llama.cpp-vulkan
 	pkgdesc = Port of Facebook's LLaMA model in C/C++ (with Vulkan GPU optimizations)
-	pkgver = b6653
+	pkgver = b6663
 	pkgrel = 1
 	url = https://github.com/ggerganov/llama.cpp
 	arch = x86_64
@@ -28,7 +28,7 @@ pkgbase = llama.cpp-vulkan
 	conflicts = stable-diffusion.cpp
 	options = lto
 	options = !debug
-	source = llama.cpp-vulkan-b6653.tar.gz::https://github.com/ggml-org/llama.cpp/archive/refs/tags/b6653.tar.gz
-	sha256sums = b00271d1617c271584696df76171340fc8ee9b02f20ac39b93a148679ae45447
+	source = llama.cpp-vulkan-b6663.tar.gz::https://github.com/ggml-org/llama.cpp/archive/refs/tags/b6663.tar.gz
+	sha256sums = 9c169965ffa4cb4dfc084af792da5f0f4537686722e735c211fb7673cb7ec569
 
 pkgname = llama.cpp-vulkan
diff --git a/PKGBUILD b/PKGBUILD
index bbb3309667585..06c16a8a976c0 100644
--- a/PKGBUILD
+++ b/PKGBUILD
@@ -3,7 +3,7 @@
 
 pkgname=llama.cpp-vulkan
 _pkgname=${pkgname%%-vulkan}
-pkgver=b6653
+pkgver=b6663
 pkgrel=1
 pkgdesc="Port of Facebook's LLaMA model in C/C++ (with Vulkan GPU optimizations)"
 arch=(x86_64 armv7h aarch64)
@@ -33,7 +33,7 @@ provides=(${_pkgname})
 conflicts=(${_pkgname} libggml ggml stable-diffusion.cpp)
 options=(lto !debug)
 source=("${pkgname}-${pkgver}.tar.gz::https://github.com/ggml-org/llama.cpp/archive/refs/tags/${pkgver}.tar.gz")
-sha256sums=('b00271d1617c271584696df76171340fc8ee9b02f20ac39b93a148679ae45447')
+sha256sums=('9c169965ffa4cb4dfc084af792da5f0f4537686722e735c211fb7673cb7ec569')
 
 prepare() {
   ln -sf "${_pkgname}-${pkgver}" llama.cpp