diff --git a/CMakeLists.txt b/CMakeLists.txt
index 8aa36e4c3..72dc2fa4b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -304,7 +304,6 @@ if (GGML_CUDA_SOURCES)
     message(STATUS "GGML CUDA sources found, configuring CUDA architecture")
     set_property(TARGET ggml PROPERTY CUDA_ARCHITECTURES OFF)
     set_property(TARGET ggml PROPERTY CUDA_SELECT_NVCC_ARCH_FLAGS "Auto")
-    set_property(TARGET llama PROPERTY CUDA_ARCHITECTURES OFF)
 endif()
 
 set(TARGET koboldcpp)
diff --git a/otherarch/gpt2_v1.cpp b/otherarch/gpt2_v1.cpp
index b70f814e9..9414725e4 100644
--- a/otherarch/gpt2_v1.cpp
+++ b/otherarch/gpt2_v1.cpp
@@ -357,7 +357,8 @@ bool legacy_gpt2_eval(
     
 
     struct ggml_v1_context * ctx0 = ggml_v1_init(params);
-    struct ggml_v1_cgraph gf = { .n_threads = n_threads };
+    struct ggml_v1_cgraph gf = {};
+    gf.n_threads = n_threads;
 
     struct ggml_v1_tensor * embd = ggml_v1_new_tensor_1d(ctx0, GGML_V1_TYPE_I32, N);
     memcpy(embd->data, embd_inp.data(), N*ggml_v1_element_size(embd));
diff --git a/otherarch/gpt2_v2.cpp b/otherarch/gpt2_v2.cpp
index 199c353b9..5f498a7a6 100644
--- a/otherarch/gpt2_v2.cpp
+++ b/otherarch/gpt2_v2.cpp
@@ -376,7 +376,8 @@ bool gpt2_eval(
     
 
     struct ggml_context * ctx0 = ggml_init(params);
-    struct ggml_cgraph gf = { .n_threads = n_threads };
+    struct ggml_cgraph gf = {};
+    gf.n_threads = n_threads;
 
     struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
     memcpy(embd->data, embd_inp.data(), N*ggml_element_size(embd));
diff --git a/otherarch/gptj_v1.cpp b/otherarch/gptj_v1.cpp
index 3ec25d2ae..67bc15062 100644
--- a/otherarch/gptj_v1.cpp
+++ b/otherarch/gptj_v1.cpp
@@ -407,7 +407,8 @@ bool legacy_gptj_eval(
     
 
     struct ggml_v1_context * ctx0 = ggml_v1_init(params);
-    struct ggml_v1_cgraph gf = { .n_threads = n_threads };
+    struct ggml_v1_cgraph gf = {};
+    gf.n_threads = n_threads;
 
     struct ggml_v1_tensor * embd = ggml_v1_new_tensor_1d(ctx0, GGML_V1_TYPE_I32, N);
     memcpy(embd->data, embd_inp.data(), N*ggml_v1_element_size(embd));
diff --git a/otherarch/gptj_v2.cpp b/otherarch/gptj_v2.cpp
index c942cd3d7..591a4bb15 100644
--- a/otherarch/gptj_v2.cpp
+++ b/otherarch/gptj_v2.cpp
@@ -388,7 +388,8 @@ bool gptj_eval(
     
 
     struct ggml_context * ctx0 = ggml_init(params);
-    struct ggml_cgraph gf = { .n_threads = n_threads };
+    struct ggml_cgraph gf = {};
+    gf.n_threads = n_threads;
 
     struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
     memcpy(embd->data, embd_inp.data(), N*ggml_element_size(embd));
diff --git a/otherarch/neox.cpp b/otherarch/neox.cpp
index 53a9d4512..02e144404 100644
--- a/otherarch/neox.cpp
+++ b/otherarch/neox.cpp
@@ -382,7 +382,8 @@ bool stablelm_eval(
     
 
     struct ggml_context * ctx0 = ggml_init(params);
-    struct ggml_cgraph gf = { .n_threads = n_threads };
+    struct ggml_cgraph gf = {};
+    gf.n_threads = n_threads;
 
     struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
     memcpy(embd->data, embd_inp.data(), N*ggml_element_size(embd));