diff --git a/include/llama.h b/include/llama.h
index 143ddb3d63..7473536f09 100644
--- a/include/llama.h
+++ b/include/llama.h
@@ -203,12 +203,12 @@ extern "C" {
     };
 
     enum llama_load_mode {
-        LLAMA_LOAD_MODE_AUTO       = 0, // auto-detect based on device capabilities
-        LLAMA_LOAD_MODE_NONE       = 1, // no special loading mode
-        LLAMA_LOAD_MODE_MMAP       = 2, // memory map the model
-        LLAMA_LOAD_MODE_MLOCK      = 3, // force system to keep model in RAM rather than swapping or compressing
-        LLAMA_LOAD_MODE_MMAP_MLOCK = 4, // mmap + force system to keep model in RAM rather than swapping or compressing
-        LLAMA_LOAD_MODE_DIRECT_IO  = 5, // use direct I/O if available
+        LLAMA_LOAD_MODE_AUTO       = -1, // auto-detect based on device capabilities
+        LLAMA_LOAD_MODE_NONE       =  0, // no special loading mode
+        LLAMA_LOAD_MODE_MMAP       =  1, // memory map the model
+        LLAMA_LOAD_MODE_MLOCK      =  2, // force system to keep model in RAM rather than swapping or compressing
+        LLAMA_LOAD_MODE_MMAP_MLOCK =  3, // mmap + force system to keep model in RAM rather than swapping or compressing
+        LLAMA_LOAD_MODE_DIRECT_IO  =  4, // use direct I/O if available
     };
 
     LLAMA_API const char * llama_load_mode_name(enum llama_load_mode load_mode);