diff --git a/examples/llama-bench/llama-bench.cpp b/examples/llama-bench/llama-bench.cpp
index dedaa34fd..8eccb8967 100644
--- a/examples/llama-bench/llama-bench.cpp
+++ b/examples/llama-bench/llama-bench.cpp
@@ -911,7 +911,7 @@ static void test_gen(llama_context * ctx, int n_gen, int n_past, int n_threads)
     }
 }
 
-static void llama_null_log_callback(enum llama_log_level level, const char * text, void * user_data) {
+static void llama_null_log_callback(enum ggml_log_level level, const char * text, void * user_data) {
     (void) level;
     (void) text;
     (void) user_data;
diff --git a/ggml-metal.h b/ggml-metal.h
index cd4531802..97d859f2c 100644
--- a/ggml-metal.h
+++ b/ggml-metal.h
@@ -19,11 +19,11 @@
 
 #pragma once
 
+#include "ggml.h"
+
 #include <stddef.h>
 #include <stdbool.h>
 
-#include "llama.h"
-
 // max memory buffers that can be mapped to the device
 #define GGML_METAL_MAX_BUFFERS 16
 #define GGML_METAL_MAX_COMMAND_BUFFERS 32
@@ -35,7 +35,7 @@ struct ggml_cgraph;
 extern "C" {
 #endif
 
-void ggml_metal_log_set_callback(void (*log_callback)(enum llama_log_level level, const char * text, void * user_data), void * user_data);
+void ggml_metal_log_set_callback(void (*log_callback)(enum ggml_log_level level, const char * text, void * user_data), void * user_data);
 
 struct ggml_metal_context;
 
diff --git a/ggml-metal.m b/ggml-metal.m
index 7ad59db33..5291d9f81 100644
--- a/ggml-metal.m
+++ b/ggml-metal.m
@@ -16,9 +16,9 @@
 #define GGML_METAL_LOG_WARN(...)
 #define GGML_METAL_LOG_ERROR(...)
 #else
-#define GGML_METAL_LOG_INFO(...)  ggml_metal_log(LLAMA_LOG_LEVEL_INFO, __VA_ARGS__)
-#define GGML_METAL_LOG_WARN(...)  ggml_metal_log(LLAMA_LOG_LEVEL_WARN, __VA_ARGS__)
-#define GGML_METAL_LOG_ERROR(...) ggml_metal_log(LLAMA_LOG_LEVEL_ERROR, __VA_ARGS__)
+#define GGML_METAL_LOG_INFO(...)  ggml_metal_log(GGML_LOG_LEVEL_INFO, __VA_ARGS__)
+#define GGML_METAL_LOG_WARN(...)  ggml_metal_log(GGML_LOG_LEVEL_WARN, __VA_ARGS__)
+#define GGML_METAL_LOG_ERROR(...) ggml_metal_log(GGML_LOG_LEVEL_ERROR, __VA_ARGS__)
 #endif
 
 #define UNUSED(x) (void)(x)
@@ -120,15 +120,15 @@ static NSString * const msl_library_source = @"see metal.metal";
 @implementation GGMLMetalClass
 @end
 
-void (*ggml_metal_log_callback)(enum llama_log_level level, const char * text, void * user_data) = NULL;
+void (*ggml_metal_log_callback)(enum ggml_log_level level, const char * text, void * user_data) = NULL;
 void *ggml_metal_log_user_data = NULL;
 
-void ggml_metal_log_set_callback(void (*log_callback)(enum llama_log_level level, const char * text, void * user_data), void * user_data) {
+void ggml_metal_log_set_callback(void (*log_callback)(enum ggml_log_level level, const char * text, void * user_data), void * user_data) {
   ggml_metal_log_callback = log_callback;
   ggml_metal_log_user_data = user_data;
 }
 
-static void ggml_metal_log(enum llama_log_level level, const char* format, ...){
+static void ggml_metal_log(enum ggml_log_level level, const char* format, ...){
   if ( ggml_metal_log_callback != NULL ) {
     va_list args;
     va_start(args, format);
diff --git a/ggml.h b/ggml.h
index c936823d6..230217664 100644
--- a/ggml.h
+++ b/ggml.h
@@ -437,6 +437,12 @@ extern "C" {
         GGML_OBJECT_WORK_BUFFER
     };
 
+    enum ggml_log_level {
+        GGML_LOG_LEVEL_ERROR = 2,
+        GGML_LOG_LEVEL_WARN = 3,
+        GGML_LOG_LEVEL_INFO = 4
+    };
+
     // ggml object
     struct ggml_object {
         size_t offs;
diff --git a/llama.cpp b/llama.cpp
index 139074993..223bbac73 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -91,15 +91,12 @@
 //
 
 LLAMA_ATTRIBUTE_FORMAT(2, 3)
-static void llama_log_internal        (enum llama_log_level level, const char* format, ...);
-static void llama_log_callback_default(enum llama_log_level level, const char * text, void * user_data);
-
-#define LLAMA_LOG_INFO(...)  llama_log_internal(LLAMA_LOG_LEVEL_INFO , __VA_ARGS__)
-#define LLAMA_LOG_WARN(...)  llama_log_internal(LLAMA_LOG_LEVEL_WARN , __VA_ARGS__)
-#define LLAMA_LOG_ERROR(...) llama_log_internal(LLAMA_LOG_LEVEL_ERROR, __VA_ARGS__)
-
-
+static void llama_log_internal        (ggml_log_level level, const char* format, ...);
+static void llama_log_callback_default(ggml_log_level level, const char * text, void * user_data);
 
+#define LLAMA_LOG_INFO(...)  llama_log_internal(GGML_LOG_LEVEL_INFO , __VA_ARGS__)
+#define LLAMA_LOG_WARN(...)  llama_log_internal(GGML_LOG_LEVEL_WARN , __VA_ARGS__)
+#define LLAMA_LOG_ERROR(...) llama_log_internal(GGML_LOG_LEVEL_ERROR, __VA_ARGS__)
 
 //
 // helpers
@@ -6370,7 +6367,7 @@ void llama_log_set(llama_log_callback log_callback, void * user_data) {
     g_state.log_callback_user_data = user_data;
 }
 
-static void llama_log_internal_v(llama_log_level level, const char * format, va_list args) {
+static void llama_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];
@@ -6387,14 +6384,14 @@ static void llama_log_internal_v(llama_log_level level, const char * format, va_
     va_end(args_copy);
 }
 
-static void llama_log_internal(llama_log_level level, const char * format, ...) {
+static void llama_log_internal(ggml_log_level level, const char * format, ...) {
     va_list args;
     va_start(args, format);
     llama_log_internal_v(level, format, args);
     va_end(args);
 }
 
-static void llama_log_callback_default(llama_log_level level, const char * text, void * user_data) {
+static void llama_log_callback_default(ggml_log_level level, const char * text, void * user_data) {
     (void) level;
     (void) user_data;
     fputs(text, stderr);
diff --git a/llama.h b/llama.h
index 37975bebe..ca5ee3a87 100644
--- a/llama.h
+++ b/llama.h
@@ -62,12 +62,6 @@ extern "C" {
 
     typedef int llama_token;
 
-    enum llama_log_level {
-        LLAMA_LOG_LEVEL_ERROR = 2,
-        LLAMA_LOG_LEVEL_WARN  = 3,
-        LLAMA_LOG_LEVEL_INFO  = 4
-    };
-
     enum llama_vocab_type {
         LLAMA_VOCAB_TYPE_SPM = 0, // SentencePiece
         LLAMA_VOCAB_TYPE_BPE = 1, // Byte Pair Encoding
@@ -156,7 +150,7 @@ extern "C" {
     // If your logging mechanism cannot handle that, check if the last character is '\n' and strip it
     // if it exists.
     // It might not exist for progress report where '.' is output repeatedly.
-    typedef void (*llama_log_callback)(enum llama_log_level level, const char * text, void * user_data);
+    typedef void (*llama_log_callback)(enum ggml_log_level level, const char * text, void * user_data);
 
     // model quantization parameters
     typedef struct llama_model_quantize_params {