kompute : disable LLAMA_SPLIT_LAYER after ggml-org#5321

cebtenzzre · cebtenzzre · commit dbf17f0b7cf8 · 2024-03-04T16:39:42.000-05:00
Signed-off-by: Jared Van Bortel &lt;jared@nomic.ai&gt;
diff --git a/llama.cpp b/llama.cpp
@@ -3763,6 +3763,7 @@ static bool llm_load_tensors(
         model.buft_layer[i] = llama_default_buffer_type_cpu(true);
     }
 
+#ifndef GGML_USE_KOMPUTE
     if (split_mode == LLAMA_SPLIT_MODE_LAYER) {
         // calculate the split points
         int device_count = llama_get_device_count();
@@ -3800,7 +3801,9 @@ static bool llm_load_tensors(
         } else {
             model.buft_output = llama_default_buffer_type_cpu(true);
         }
-    } else {
+    } else
+#endif
+    {
         ggml_backend_buffer_type_t split_buft;
         if (split_mode == LLAMA_SPLIT_MODE_ROW) {
             split_buft = llama_default_buffer_type_split(main_gpu, tensor_split);
