examples : fix CUBLAS leftovers (#0)

ggml-ci

diff --git a/examples/gpt-2/main-backend.cpp b/examples/gpt-2/main-backend.cpp
index bd4cb80b17d47d4d23bb30e61c31207945240614..714c158fb206a4168dd5cb089da0a548b648d133 100644 (file)
--- a/examples/gpt-2/main-backend.cpp
+++ b/examples/gpt-2/main-backend.cpp
@@ -2,7 +2,7 @@
 #include "ggml/ggml-alloc.h"
 #include "ggml/ggml-backend.h"
 
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
 #include "ggml-cuda.h"
 #endif
 
@@ -197,7 +197,7 @@ bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab &
     }
 
     // initialize the backend
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
     if (n_gpu_layers > 0) {
         fprintf(stderr, "%s: using CUDA backend\n", __func__);
         model.backend = ggml_backend_cuda_init(0);

diff --git a/examples/gpt-2/main-batched.cpp b/examples/gpt-2/main-batched.cpp
index 9ba4496cf10357376739d4005d9946a4b38fd9af..6dbf5e3bc119360b645196ee489e212a2ffe1490 100644 (file)
--- a/examples/gpt-2/main-batched.cpp
+++ b/examples/gpt-2/main-batched.cpp
@@ -2,7 +2,7 @@
 #include "ggml/ggml-alloc.h"
 #include "ggml/ggml-backend.h"
 
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
 #include "ggml-cuda.h"
 #endif
 
@@ -285,7 +285,7 @@ bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab &
     }
 
     // initialize the backend
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
     if (n_gpu_layers > 0) {
         fprintf(stderr, "%s: using CUDA backend\n", __func__);
         model.backend = ggml_backend_cuda_init(0);

diff --git a/examples/gpt-2/main-sched.cpp b/examples/gpt-2/main-sched.cpp
index ad10aa96125bc8722041029f1eb8b5300d2a5d99..bdf3bff8233cbfc061279c0469868ef816462b74 100644 (file)
--- a/examples/gpt-2/main-sched.cpp
+++ b/examples/gpt-2/main-sched.cpp
@@ -2,7 +2,7 @@
 #include "ggml/ggml-alloc.h"
 #include "ggml/ggml-backend.h"
 
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
 #include "ggml-cuda.h"
 #endif
 
@@ -105,7 +105,7 @@ void init_backends(gpt2_model & model, const gpt_params & params) {
     ggml_backend_t gpu_backend = NULL;
 
     // initialize the backends
-#ifdef GGML_USE_CUBLAS
+#ifdef GGML_USE_CUDA
     if (params.n_gpu_layers > 0) {
         fprintf(stderr, "%s: using CUDA backend\n", __func__);
         gpu_backend = ggml_backend_cuda_init(0);

diff --git a/examples/python/README.md b/examples/python/README.md
index 480920f74cb28abee40027d5a48c74df609b5df6..69287f88f3f7a96d0a3b35488bf0d02647def769 100644 (file)
--- a/examples/python/README.md
+++ b/examples/python/README.md
@@ -52,7 +52,7 @@ As of this writing the best is to use [ggerganov/llama.cpp](https://github.com/g
 
 ```bash
 git clone https://github.com/ggerganov/llama.cpp
-# On a CUDA-enabled system add -DLLAMA_CUBLAS=1
+# On a CUDA-enabled system add -DLLAMA_CUDA=1
 # On a Mac add -DLLAMA_METAL=1
 cmake llama.cpp \
   -B llama_build \

diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index e13733f2e0b09ee194905652d3abecc587c0be57..47b7a2d93dde7f5539c055f05ddedc9569120c77 100644 (file)
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -215,7 +215,7 @@ if (GGML_CUDA)
 
     find_package(CUDAToolkit)
     if (CUDAToolkit_FOUND)
-        message(STATUS "cuBLAS found")
+        message(STATUS "CUDA found")
 
         enable_language(CUDA)
 
@@ -252,10 +252,11 @@ if (GGML_CUDA)
             set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo")
         endif()
     else()
-        message(WARNING "cuBLAS not found")
+        message(WARNING "CUDA not found")
     endif()
 endif()
 
+# TODO: do not build separate ggml-rocm target (see CUDA build above, or llama.cpp for reference)
 if (GGML_HIPBLAS)
     list(APPEND CMAKE_PREFIX_PATH /opt/rocm)