diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h
index ce3d92cb2..57b44b85c 100644
--- a/ggml/include/ggml.h
+++ b/ggml/include/ggml.h
@@ -2177,6 +2177,10 @@ extern "C" {
     typedef void (*ggml_opt_callback)(void * data, int accum_step, float * sched, bool * cancel);
     typedef void (*ggml_log_callback)(enum ggml_log_level level, const char * text, void * user_data);
 
+    // Set callback for all future logging events.
+    // If this is not called, or NULL is supplied, everything is output on stderr.
+    GGML_API void ggml_log_set(ggml_log_callback log_callback, void * user_data);
+
     // optimization parameters
     //
     //   see ggml.c (ggml_opt_default_params) for default values
diff --git a/ggml/src/ggml-impl.h b/ggml/src/ggml-impl.h
index 833984190..b906ecb38 100644
--- a/ggml/src/ggml-impl.h
+++ b/ggml/src/ggml-impl.h
@@ -33,6 +33,31 @@ extern "C" {
 #endif
 #endif
 
+//
+// logging
+//
+
+#ifdef __GNUC__
+#ifdef __MINGW32__
+#define GGML_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__)))
+#else
+#define GGML_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__)))
+#endif
+#else
+#define GGML_ATTRIBUTE_FORMAT(...)
+#endif
+
+GGML_ATTRIBUTE_FORMAT(2, 3)
+void ggml_log_internal        (ggml_log_level level, const char * format, ...);
+void ggml_log_callback_default(ggml_log_level level, const char * text, void * user_data);
+
+#define GGML_LOG(...)       ggml_log_internal(GGML_LOG_LEVEL_NONE , __VA_ARGS__)
+#define GGML_LOG_INFO(...)  ggml_log_internal(GGML_LOG_LEVEL_INFO , __VA_ARGS__)
+#define GGML_LOG_WARN(...)  ggml_log_internal(GGML_LOG_LEVEL_WARN , __VA_ARGS__)
+#define GGML_LOG_ERROR(...) ggml_log_internal(GGML_LOG_LEVEL_ERROR, __VA_ARGS__)
+#define GGML_LOG_DEBUG(...) ggml_log_internal(GGML_LOG_LEVEL_DEBUG, __VA_ARGS__)
+#define GGML_LOG_CONT(...)  ggml_log_internal(GGML_LOG_LEVEL_CONT , __VA_ARGS__)
+
 // bitset
 
 typedef uint32_t ggml_bitset_t;
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index bcbc32d91..ebf35d29f 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -319,6 +319,43 @@ void ggml_abort(const char * file, int line, const char * fmt, ...) {
 // logging
 //
 
+struct ggml_logger_state {
+    ggml_log_callback log_callback = ggml_log_callback_default;
+    void * log_callback_user_data  = nullptr;
+};
+static ggml_logger_state g_logger_state;
+
+static void ggml_log_internal_v(ggml_log_level level, const char * format, va_list args) {
+    va_list args_copy;
+    va_copy(args_copy, args);
+    char buffer[128];
+    int len = vsnprintf(buffer, 128, format, args);
+    if (len < 128) {
+        g_logger_state.log_callback(level, buffer, g_logger_state.log_callback_user_data);
+    } else {
+        char * buffer2 = new char[len + 1];
+        vsnprintf(buffer2, len + 1, format, args_copy);
+        buffer2[len] = 0;
+        g_logger_state.log_callback(level, buffer2, g_logger_state.log_callback_user_data);
+        delete[] buffer2;
+    }
+    va_end(args_copy);
+}
+
+void ggml_log_internal(ggml_log_level level, const char * format, ...) {
+    va_list args;
+    va_start(args, format);
+    ggml_log_internal_v(level, format, args);
+    va_end(args);
+}
+
+void ggml_log_callback_default(ggml_log_level level, const char * text, void * user_data) {
+    (void) level;
+    (void) user_data;
+    fputs(text, stderr);
+    fflush(stderr);
+}
+
 #if (GGML_DEBUG >= 1)
 #define GGML_PRINT_DEBUG(...) printf(__VA_ARGS__)
 #else
@@ -23237,4 +23274,9 @@ int ggml_cpu_get_sve_cnt(void) {
     return 0;
 #endif
 }
+
+void ggml_log_set(ggml_log_callback log_callback, void * user_data) {
+    g_logger_state.log_callback = log_callback ? log_callback : ggml_log_callback_default;
+    g_logger_state.log_callback_user_data = user_data;
+}
 ////////////////////////////////////////////////////////////////////////////////
diff --git a/src/llama.cpp b/src/llama.cpp
index 4c0a1bb61..dac1a17a1 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -2301,22 +2301,11 @@ static ggml_backend_buffer_type_t llama_default_buffer_type_cpu(bool host_buffer
 //
 
 struct llama_state {
-    llama_state() {
-#ifdef GGML_USE_METAL
-        ggml_backend_metal_log_set_callback(log_callback, log_callback_user_data);
-#elif defined(GGML_USE_CUDA)
-        ggml_backend_cuda_log_set_callback(log_callback, log_callback_user_data);
-#elif defined(GGML_USE_CANN)
-        ggml_backend_cann_log_set_callback(log_callback, log_callback_user_data);
-#endif
-    }
-
-    // We save the log callback globally
     ggml_log_callback log_callback = llama_log_callback_default;
     void * log_callback_user_data = nullptr;
 };
 
-static llama_state g_state;
+static llama_state g_logger_state;
 
 // available llama models
 enum e_model {
@@ -21772,15 +21761,9 @@ const std::vector<std::pair<std::string, struct ggml_tensor *>> & llama_internal
 }
 
 void llama_log_set(ggml_log_callback log_callback, void * user_data) {
-    g_state.log_callback = log_callback ? log_callback : llama_log_callback_default;
-    g_state.log_callback_user_data = user_data;
-#ifdef GGML_USE_METAL
-    ggml_backend_metal_log_set_callback(g_state.log_callback, g_state.log_callback_user_data);
-#elif defined(GGML_USE_CUDA)
-    ggml_backend_cuda_log_set_callback(g_state.log_callback, g_state.log_callback_user_data);
-#elif defined(GGML_USE_CANN)
-    ggml_backend_cann_log_set_callback(g_state.log_callback, g_state.log_callback_user_data);
-#endif
+    ggml_log_set(log_callback, user_data);
+    g_logger_state.log_callback = log_callback ? log_callback : llama_log_callback_default;
+    g_logger_state.log_callback_user_data = user_data;
 }
 
 static void llama_log_internal_v(ggml_log_level level, const char * format, va_list args) {
@@ -21789,12 +21772,12 @@ static void llama_log_internal_v(ggml_log_level level, const char * format, va_l
     char buffer[128];
     int len = vsnprintf(buffer, 128, format, args);
     if (len < 128) {
-        g_state.log_callback(level, buffer, g_state.log_callback_user_data);
+        g_logger_state.log_callback(level, buffer, g_logger_state.log_callback_user_data);
     } else {
         char * buffer2 = new char[len + 1];
         vsnprintf(buffer2, len + 1, format, args_copy);
         buffer2[len] = 0;
-        g_state.log_callback(level, buffer2, g_state.log_callback_user_data);
+        g_logger_state.log_callback(level, buffer2, g_logger_state.log_callback_user_data);
         delete[] buffer2;
     }
     va_end(args_copy);