ggml : fix Q4_3 cuBLAS

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 2d4e30e5a2dea9861cfd8ef7f4d8e9ce23de1339..11ebe9eb66fae1f21e99f33616aca11206f96c66 100644 (file)
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -312,6 +312,7 @@ add_library(ggml OBJECT
 target_include_directories(ggml PUBLIC .)
 target_compile_features(ggml PUBLIC c_std_11) # don't bump
 target_link_libraries(ggml PUBLIC Threads::Threads ${LLAMA_EXTRA_LIBS})
+
 if (BUILD_SHARED_LIBS)
     set_target_properties(ggml PROPERTIES POSITION_INDEPENDENT_CODE ON)
 endif()
@@ -324,6 +325,7 @@ add_library(llama
 target_include_directories(llama PUBLIC .)
 target_compile_features(llama PUBLIC cxx_std_11) # don't bump
 target_link_libraries(llama PRIVATE ggml ${LLAMA_EXTRA_LIBS})
+
 if (BUILD_SHARED_LIBS)
     set_target_properties(llama PROPERTIES POSITION_INDEPENDENT_CODE ON)
     target_compile_definitions(llama PRIVATE LLAMA_SHARED LLAMA_BUILD)

diff --git a/ggml.c b/ggml.c
index d9a95af8d4b5c279ff3d8a195141b80c800037c4..281b20283c16f858f53bf506fccae122ea0a2f20 100644 (file)
--- a/ggml.c
+++ b/ggml.c
@@ -7992,6 +7992,9 @@ static void ggml_compute_forward_mul_mat_q_f32(
         else if (type == GGML_TYPE_Q4_2) {
             dequantize_row_q_cuda = dequantize_row_q4_2_cuda;
         }
+        else if (type == GGML_TYPE_Q4_3) {
+            dequantize_row_q_cuda = dequantize_row_q4_3_cuda;
+        }
         else {
             GGML_ASSERT(false);
         }