CUBLAS -> CUDA

Author: dridri

CMakeLists.txt

@@ -7,10 +7,10 @@ option(DEBUG "Debug mode" OFF)
 option(GGML_CUDA "cuda mode" OFF)
 
 if(DEBUG)
-    set( CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -g3 -O0" )
-    set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g3 -O0" )
-    set( CMAKE_LD_FLAGS "${CMAKE_LD_FLAGS} -g3 -O0" )
-    set( CMAKE_LINKER_FLAGS "${CMAKE_LINKER_FLAGS} -g3 -O0" )
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -g3 -O0")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g3 -O0")
+    set(CMAKE_LD_FLAGS "${CMAKE_LD_FLAGS} -g3 -O0")
+    set(CMAKE_LINKER_FLAGS "${CMAKE_LINKER_FLAGS} -g3 -O0")
     set(GGML_DEBUG ON CACHE BOOL "Debug mode")
 endif()
 

README.md

@@ -23,7 +23,7 @@ This is tested with mac os arm
 ````
 mkdir build
 cd build
-cmake .. -DGGML_CUBLAS=ON
+cmake .. -DGGML_CUDA=ON
 make
 ````
 This is tested with Ubuntu 22.04 and cuda 12.0 and a 1070ti

main.cpp

@@ -2,7 +2,7 @@
 #include "ggml/include/ggml-backend.h"
 #include "ggml/include/ggml.h"
 
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
 #include "ggml-cuda.h"
 #endif
 
@@ -612,7 +612,7 @@ bool autoregressive_model_load(const std::string &fname,
   }
 
   // initialize the backend
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
   fprintf(stderr, "%s: using CUDA backend\n", __func__);
   model.backend = ggml_backend_cuda_init(0);
   if (!model.backend) {
@@ -1074,7 +1074,7 @@ bool diffusion_model_load(const std::string &fname, diffusion_model &model) {
   }
 
   // initialize the backend
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
   fprintf(stderr, "%s: using CUDA backend\n", __func__);
   model.backend = ggml_backend_cuda_init(0);
   if (!model.backend) {
@@ -1609,7 +1609,7 @@ bool vocoder_model_load(const std::string &fname, vocoder_model &model) {
   }
 
   // initialize the backend
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
   fprintf(stderr, "%s: using CUDA backend\n", __func__);
   model.backend = ggml_backend_cuda_init(0);
   if (!model.backend) {